antonlyap/bliss-rs
1
0
Fork 0
Code Issues Pull Requests Packages Activity

Compare commits

base: antonlyap:75d6db0ee8bd5d596ff945b32854219fe12b2624
Branches Tags
antonlyap:master
antonlyap:make-ffmpeg-optional
antonlyap:add-rasperry-pi
antonlyap:bliss-python-support
antonlyap:fix-tag-bug
antonlyap:disable-parallel
antonlyap:minor-fixes-2
antonlyap:update-changlog
antonlyap:change-analysis
antonlyap:v0.0.4
antonlyap:v0.0.3
...
compare: antonlyap:f446e4ade73bddd8c6793704aa3fe1241a6663bc
Branches Tags
antonlyap:master
antonlyap:make-ffmpeg-optional
antonlyap:add-rasperry-pi
antonlyap:bliss-python-support
antonlyap:fix-tag-bug
antonlyap:disable-parallel
antonlyap:minor-fixes-2
antonlyap:update-changlog
antonlyap:change-analysis
antonlyap:v0.0.4
antonlyap:v0.0.3

2 Commits
75d6db0ee8 ... f446e4ade7

Author SHA1 Message Date
Anton
f446e4ade7 Remove path and duration from the decoded song 2024-02-16 17:09:20 +02:00
Anton
719f6baaa4 Remove unused stuff 2024-02-11 09:20:48 +02:00
12 changed files with 100 additions and 6490 deletions
Show Stats Expand all files Collapse all files

838
Cargo.lock generated
View File

File diff suppressed because it is too large Load Diff

45
Cargo.toml
View File

@ -19,36 +19,14 @@ crate-type = ["rlib", "cdylib"]
features = ["bliss-audio-aubio-rs/rustdoc", "library"]
no-default-features = true
[features]
default = ["bliss-audio-aubio-rs/static"]
# Build ffmpeg instead of using the host's.
build-ffmpeg = ["ffmpeg-next/build"]
ffmpeg-static = ["ffmpeg-next/static"]
# Build for raspberry pis
rpi = ["ffmpeg-next/rpi"]
# Use if you get "No prebuilt bindings. Try use `bindgen` feature"
update-aubio-bindings = ["bliss-audio-aubio-rs/bindgen"]
# Use if you want to build python bindings with maturin.
python-bindings = ["bliss-audio-aubio-rs/fftw3"]
# Enable the benchmarks with `cargo +nightly bench --features=bench`
bench = []
library = [
"serde", "dep:rusqlite", "dep:dirs", "dep:tempdir",
"dep:anyhow", "dep:serde_ini", "dep:serde_json",
"dep:indicatif",
]
serde = ["dep:serde"]
[dependencies]
# Until https://github.com/aubio/aubio/issues/336 is somehow solved
# Hopefully we'll be able to use the official aubio-rs at some point.
bliss-audio-aubio-rs = { version = "0.2.1", features = ["bindgen"] }
bliss-audio-aubio-rs = { version = "0.2.1", features = ["static"] }
crossbeam = "0.8.2"
ffmpeg-next = "6.0.0"
ffmpeg-sys-next = { version = "6.0.1", default-features = false }
ffmpeg-next = { version = "6.0.0", features = ["static"] }
log = "0.4.17"
ndarray = { version = "0.15.6", features = ["rayon"] }
num_cpus = "1.15.0"
ndarray-stats = "0.5.1"
noisy_float = "0.2.0"
ripemd = "0.1.3"
@ -56,17 +34,11 @@ rustfft = "6.1.0"
thiserror = "1.0.40"
strum = "0.24.1"
strum_macros = "0.24.3"
rcue = "0.1.3"
# Deps for the library feature
serde = { version = "1.0", optional = true, features = ["derive"] }
serde_json = { version = "1.0.59", optional = true }
serde_ini = { version = "0.2.0", optional = true }
tempdir = { version = "0.3.7", optional = true }
rusqlite = { version = "0.28.0", optional = true }
dirs = { version = "5.0.0", optional = true }
anyhow = { version = "1.0.58", optional = true }
indicatif = { version = "0.17.0", optional = true }
[dependencies.neon]
version = "1.0.0-alpha.4"
@ -81,16 +53,3 @@ anyhow = "1.0.45"
clap = "2.33.3"
pretty_assertions = "1.3.0"
serde_json = "1.0.59"
[[example]]
name = "library"
required-features = ["library"]
[[example]]
name = "library_extra_info"
required-features = ["library"]
[[example]]
name = "playlist"
required-features = ["serde"]

9
examples/distance.rs
View File

@ -16,11 +16,18 @@ fn main() -> Result<(), String> {
let song1 = Song::from_path(first_path).map_err(|x| x.to_string())?;
let song2 = Song::from_path(second_path).map_err(|x| x.to_string())?;
let mut distance_squared: f64 = 0.0;
let analysis1 = song1.analysis.as_bytes();
let analysis2 = song2.analysis.as_bytes();
for (i, feature1) in analysis1.iter().enumerate() {
distance_squared += (feature1 - analysis2[i]).pow(2) as f64;
}
println!(
"d({:?}, {:?}) = {}",
song1.path,
song2.path,
song1.distance(&song2)
distance_squared.sqrt(),
);
Ok(())
}

204
examples/library.rs
View File

@ -1,204 +0,0 @@
/// Basic example of how one would combine bliss with an "audio player",
/// through [Library].
///
/// For simplicity's sake, this example recursively gets songs from a folder
/// to emulate an audio player library, without handling CUE files.
use anyhow::Result;
use bliss_audio::library::{AppConfigTrait, BaseConfig, Library};
use clap::{App, Arg, SubCommand};
use glob::glob;
use serde::{Deserialize, Serialize};
use std::fs;
use std::num::NonZeroUsize;
use std::path::{Path, PathBuf};
#[derive(Serialize, Deserialize, Clone, Debug)]
// A config structure, that will be serialized as a
// JSON file upon Library creation.
pub struct Config {
#[serde(flatten)]
// The base configuration, containing both the config file
// path, as well as the database path.
pub base_config: BaseConfig,
// An extra field, to store the music library path. Any number
// of arbitrary fields (even Serializable structures) can
// of course be added.
pub music_library_path: PathBuf,
}
impl Config {
pub fn new(
music_library_path: PathBuf,
config_path: Option<PathBuf>,
database_path: Option<PathBuf>,
number_cores: Option<NonZeroUsize>,
) -> Result<Self> {
let base_config = BaseConfig::new(config_path, database_path, number_cores)?;
Ok(Self {
base_config,
music_library_path,
})
}
}
// The AppConfigTrait must know how to access the base config.
impl AppConfigTrait for Config {
fn base_config(&self) -> &BaseConfig {
&self.base_config
}
fn base_config_mut(&mut self) -> &mut BaseConfig {
&mut self.base_config
}
}
// A trait allowing to implement methods for the Library,
// useful if you don't need to store extra information in fields.
// Otherwise, doing
// ```
// struct CustomLibrary {
// library: Library<Config>,
// extra_field: ...,
// }
// ```
// and implementing functions for that struct would be the way to go.
// That's what the [reference](https://github.com/Polochon-street/blissify-rs)
// implementation does.
trait CustomLibrary {
fn song_paths(&self) -> Result<Vec<String>>;
}
impl CustomLibrary for Library<Config> {
/// Get all songs in the player library
fn song_paths(&self) -> Result<Vec<String>> {
let music_path = &self.config.music_library_path;
let pattern = Path::new(&music_path).join("**").join("*");
Ok(glob(&pattern.to_string_lossy())?
.map(|e| fs::canonicalize(e.unwrap()).unwrap())
.filter(|e| match mime_guess::from_path(e).first() {
Some(m) => m.type_() == "audio",
None => false,
})
.map(|x| x.to_string_lossy().to_string())
.collect::<Vec<String>>())
}
}
// A simple example of what a CLI-app would look.
//
// Note that `Library::new` is used only on init, and subsequent
// commands use `Library::from_path`.
fn main() -> Result<()> {
let matches = App::new("library-example")
.version(env!("CARGO_PKG_VERSION"))
.author("Polochon_street")
.about("Example binary implementing bliss for an audio player.")
.subcommand(
SubCommand::with_name("init")
.about(
"Initialize a Library, both storing the config and analyzing folders
containing songs.",
)
.arg(
Arg::with_name("FOLDER")
.help("A folder containing the music library to analyze.")
.required(true),
)
.arg(
Arg::with_name("database-path")
.short("d")
.long("database-path")
.help(
"Optional path where to store the database file containing
the songs' analysis. Defaults to XDG_DATA_HOME/bliss-rs/bliss.db.",
)
.takes_value(true),
)
.arg(
Arg::with_name("config-path")
.short("c")
.long("config-path")
.help(
"Optional path where to store the config file containing
the library setup. Defaults to XDG_DATA_HOME/bliss-rs/config.json.",
)
.takes_value(true),
),
)
.subcommand(
SubCommand::with_name("update")
.about(
"Update a Library's songs, trying to analyze failed songs,
as well as songs not in the library.",
)
.arg(
Arg::with_name("config-path")
.short("c")
.long("config-path")
.help(
"Optional path where to load the config file containing
the library setup. Defaults to XDG_DATA_HOME/bliss-rs/config.json.",
)
.takes_value(true),
),
)
.subcommand(
SubCommand::with_name("playlist")
.about(
"Make a playlist, starting with the song at SONG_PATH, returning
the songs' paths.",
)
.arg(Arg::with_name("SONG_PATH").takes_value(true))
.arg(
Arg::with_name("config-path")
.short("c")
.long("config-path")
.help(
"Optional path where to load the config file containing
the library setup. Defaults to XDG_DATA_HOME/bliss-rs/config.json.",
)
.takes_value(true),
)
.arg(
Arg::with_name("playlist-length")
.short("l")
.long("playlist-length")
.help("Optional playlist length. Defaults to 20.")
.takes_value(true),
),
)
.get_matches();
if let Some(sub_m) = matches.subcommand_matches("init") {
let folder = PathBuf::from(sub_m.value_of("FOLDER").unwrap());
let config_path = sub_m.value_of("config-path").map(PathBuf::from);
let database_path = sub_m.value_of("database-path").map(PathBuf::from);
let config = Config::new(folder, config_path, database_path, None)?;
let mut library = Library::new(config)?;
library.analyze_paths(library.song_paths()?, true)?;
} else if let Some(sub_m) = matches.subcommand_matches("update") {
let config_path = sub_m.value_of("config-path").map(PathBuf::from);
let mut library: Library<Config> = Library::from_config_path(config_path)?;
library.update_library(library.song_paths()?, true, true)?;
} else if let Some(sub_m) = matches.subcommand_matches("playlist") {
let song_path = sub_m.value_of("SONG_PATH").unwrap();
let config_path = sub_m.value_of("config-path").map(PathBuf::from);
let playlist_length = sub_m
.value_of("playlist-length")
.unwrap_or("20")
.parse::<usize>()?;
let library: Library<Config> = Library::from_config_path(config_path)?;
let songs = library.playlist_from::<()>(song_path, playlist_length)?;
let song_paths = songs
.into_iter()
.map(|s| s.bliss_song.path.to_string_lossy().to_string())
.collect::<Vec<String>>();
for song in song_paths {
println!("{song:?}");
}
}
Ok(())
}

227
examples/library_extra_info.rs
View File

@ -1,227 +0,0 @@
/// Basic example of how one would combine bliss with an "audio player",
/// through [Library], showing how to put extra info in the database for
/// each song.
///
/// For simplicity's sake, this example recursively gets songs from a folder
/// to emulate an audio player library, without handling CUE files.
use anyhow::Result;
use bliss_audio::library::{AppConfigTrait, BaseConfig, Library};
use clap::{App, Arg, SubCommand};
use glob::glob;
use serde::{Deserialize, Serialize};
use std::fs;
use std::num::NonZeroUsize;
use std::path::{Path, PathBuf};
#[derive(Serialize, Deserialize, Clone, Debug)]
/// A config structure, that will be serialized as a
/// JSON file upon Library creation.
pub struct Config {
#[serde(flatten)]
/// The base configuration, containing both the config file
/// path, as well as the database path.
pub base_config: BaseConfig,
/// An extra field, to store the music library path. Any number
/// of arbitrary fields (even Serializable structures) can
/// of course be added.
pub music_library_path: PathBuf,
}
impl Config {
pub fn new(
music_library_path: PathBuf,
config_path: Option<PathBuf>,
database_path: Option<PathBuf>,
number_cores: Option<NonZeroUsize>,
) -> Result<Self> {
let base_config = BaseConfig::new(config_path, database_path, number_cores)?;
Ok(Self {
base_config,
music_library_path,
})
}
}
// The AppConfigTrait must know how to access the base config.
impl AppConfigTrait for Config {
fn base_config(&self) -> &BaseConfig {
&self.base_config
}
fn base_config_mut(&mut self) -> &mut BaseConfig {
&mut self.base_config
}
}
// A trait allowing to implement methods for the Library,
// useful if you don't need to store extra information in fields.
// Otherwise, doing
// ```
// struct CustomLibrary {
// library: Library<Config>,
// extra_field: ...,
// }
// ```
// and implementing functions for that struct would be the way to go.
// That's what the [reference](https://github.com/Polochon-street/blissify-rs)
// implementation does.
trait CustomLibrary {
fn song_paths_info(&self) -> Result<Vec<(String, ExtraInfo)>>;
}
impl CustomLibrary for Library<Config> {
/// Get all songs in the player library, along with the extra info
/// one would want to store along with each song.
fn song_paths_info(&self) -> Result<Vec<(String, ExtraInfo)>> {
let music_path = &self.config.music_library_path;
let pattern = Path::new(&music_path).join("**").join("*");
Ok(glob(&pattern.to_string_lossy())?
.map(|e| fs::canonicalize(e.unwrap()).unwrap())
.filter_map(|e| {
mime_guess::from_path(&e).first().map(|m| {
(
e.to_string_lossy().to_string(),
ExtraInfo {
extension: e.extension().map(|e| e.to_string_lossy().to_string()),
file_name: e.file_name().map(|e| e.to_string_lossy().to_string()),
mime_type: format!("{}/{}", m.type_(), m.subtype()),
},
)
})
})
.collect::<Vec<(String, ExtraInfo)>>())
}
}
#[derive(Deserialize, Serialize, Debug, PartialEq, Clone, Default)]
// An (somewhat simple) example of what extra metadata one would put, along
// with song analysis data.
struct ExtraInfo {
extension: Option<String>,
file_name: Option<String>,
mime_type: String,
}
// A simple example of what a CLI-app would look.
//
// Note that `Library::new` is used only on init, and subsequent
// commands use `Library::from_path`.
fn main() -> Result<()> {
let matches = App::new("library-example")
.version(env!("CARGO_PKG_VERSION"))
.author("Polochon_street")
.about("Example binary implementing bliss for an audio player.")
.subcommand(
SubCommand::with_name("init")
.about(
"Initialize a Library, both storing the config and analyzing folders
containing songs.",
)
.arg(
Arg::with_name("FOLDER")
.help("A folder containing the music library to analyze.")
.required(true),
)
.arg(
Arg::with_name("database-path")
.short("d")
.long("database-path")
.help(
"Optional path where to store the database file containing
the songs' analysis. Defaults to XDG_DATA_HOME/bliss-rs/bliss.db.",
)
.takes_value(true),
)
.arg(
Arg::with_name("config-path")
.short("c")
.long("config-path")
.help(
"Optional path where to store the config file containing
the library setup. Defaults to XDG_DATA_HOME/bliss-rs/config.json.",
)
.takes_value(true),
),
)
.subcommand(
SubCommand::with_name("update")
.about(
"Update a Library's songs, trying to analyze failed songs,
as well as songs not in the library.",
)
.arg(
Arg::with_name("config-path")
.short("c")
.long("config-path")
.help(
"Optional path where to load the config file containing
the library setup. Defaults to XDG_DATA_HOME/bliss-rs/config.json.",
)
.takes_value(true),
),
)
.subcommand(
SubCommand::with_name("playlist")
.about(
"Make a playlist, starting with the song at SONG_PATH, returning
the songs' paths.",
)
.arg(Arg::with_name("SONG_PATH").takes_value(true))
.arg(
Arg::with_name("config-path")
.short("c")
.long("config-path")
.help(
"Optional path where to load the config file containing
the library setup. Defaults to XDG_DATA_HOME/bliss-rs/config.json.",
)
.takes_value(true),
)
.arg(
Arg::with_name("playlist-length")
.short("l")
.long("playlist-length")
.help("Optional playlist length. Defaults to 20.")
.takes_value(true),
),
)
.get_matches();
if let Some(sub_m) = matches.subcommand_matches("init") {
let folder = PathBuf::from(sub_m.value_of("FOLDER").unwrap());
let config_path = sub_m.value_of("config-path").map(PathBuf::from);
let database_path = sub_m.value_of("database-path").map(PathBuf::from);
let config = Config::new(folder, config_path, database_path, None)?;
let mut library = Library::new(config)?;
library.analyze_paths_extra_info(library.song_paths_info()?, true)?;
} else if let Some(sub_m) = matches.subcommand_matches("update") {
let config_path = sub_m.value_of("config-path").map(PathBuf::from);
let mut library: Library<Config> = Library::from_config_path(config_path)?;
library.update_library_extra_info(library.song_paths_info()?, true, true)?;
} else if let Some(sub_m) = matches.subcommand_matches("playlist") {
let song_path = sub_m.value_of("SONG_PATH").unwrap();
let config_path = sub_m.value_of("config-path").map(PathBuf::from);
let playlist_length = sub_m
.value_of("playlist-length")
.unwrap_or("20")
.parse::<usize>()?;
let library: Library<Config> = Library::from_config_path(config_path)?;
let songs = library.playlist_from::<ExtraInfo>(song_path, playlist_length)?;
let playlist = songs
.into_iter()
.map(|s| {
(
s.bliss_song.path.to_string_lossy().to_string(),
s.extra_info.mime_type,
)
})
.collect::<Vec<(String, String)>>();
for (path, mime_type) in playlist {
println!("{path} <{mime_type}>");
}
}
Ok(())
}

95
examples/playlist.rs
View File

@ -1,95 +0,0 @@
use anyhow::Result;
use bliss_rs::playlist::{closest_to_first_song, dedup_playlist, euclidean_distance};
use bliss_rs::bliss_lib::{analyze_paths, Song};
use clap::{App, Arg};
use glob::glob;
use std::env;
use std::fs;
use std::io::BufReader;
use std::path::{Path, PathBuf};
/* Analyzes a folder recursively, and make a playlist out of the file
* provided by the user. */
// How to use: ./playlist [-o file.m3u] [-a analysis.json] <folder> <file to start the playlist from>
fn main() -> Result<()> {
let matches = App::new("playlist")
.version(env!("CARGO_PKG_VERSION"))
.author("Polochon_street")
.about("Analyze a folder and make a playlist from a target song")
.arg(Arg::with_name("output-playlist").short("o").long("output-playlist")
.value_name("PLAYLIST.M3U")
.help("Outputs the playlist to a file.")
.takes_value(true))
.arg(Arg::with_name("analysis-file").short("a").long("analysis-file")
.value_name("ANALYSIS.JSON")
.help("Use the songs that have been analyzed in <analysis-file>, and appends newly analyzed songs to it. Defaults to /tmp/analysis.json.")
.takes_value(true))
.arg(Arg::with_name("FOLDER").help("Folders containing some songs.").required(true))
.arg(Arg::with_name("FIRST-SONG").help("Song to start from (can be outside of FOLDER).").required(true))
.get_matches();
let folder = matches.value_of("FOLDER").unwrap();
let file = fs::canonicalize(matches.value_of("FIRST-SONG").unwrap())?;
let pattern = Path::new(folder).join("**").join("*");
let mut songs: Vec<Song> = Vec::new();
let analysis_path = matches
.value_of("analysis-file")
.unwrap_or("/tmp/analysis.json");
let analysis_file = fs::File::open(analysis_path);
if let Ok(f) = analysis_file {
let reader = BufReader::new(f);
songs = serde_json::from_reader(reader)?;
}
let analyzed_paths = songs
.iter()
.map(|s| s.path.to_owned())
.collect::<Vec<PathBuf>>();
let paths = glob(&pattern.to_string_lossy())?
.map(|e| fs::canonicalize(e.unwrap()).unwrap())
.filter(|e| match mime_guess::from_path(e).first() {
Some(m) => m.type_() == "audio",
None => false,
})
.map(|x| x.to_string_lossy().to_string())
.collect::<Vec<String>>();
let song_iterator = analyze_paths(
paths
.iter()
.filter(|p| !analyzed_paths.contains(&PathBuf::from(p)))
.map(|p| p.to_owned())
.collect::<Vec<String>>(),
);
let first_song = Song::from_path(file)?;
let mut analyzed_songs = vec![first_song.to_owned()];
for (path, result) in song_iterator {
match result {
Ok(song) => analyzed_songs.push(song),
Err(e) => println!("error analyzing {}: {}", path.display(), e),
};
}
analyzed_songs.extend_from_slice(&songs);
let serialized = serde_json::to_string(&analyzed_songs).unwrap();
let mut songs_to_chose_from: Vec<_> = analyzed_songs
.into_iter()
.filter(|x| x == &first_song || paths.contains(&x.path.to_string_lossy().to_string()))
.collect();
closest_to_first_song(&first_song, &mut songs_to_chose_from, euclidean_distance);
dedup_playlist(&mut songs_to_chose_from, None);
fs::write(analysis_path, serialized)?;
let playlist = songs_to_chose_from
.iter()
.map(|s| s.path.to_string_lossy().to_string())
.collect::<Vec<String>>()
.join("\n");
if let Some(m) = matches.value_of("output-playlist") {
fs::write(m, playlist)?;
} else {
println!("{playlist}");
}
Ok(())
}

240
src/bliss_lib.rs
View File

@ -15,14 +15,6 @@
//! is as easy as computing distances between that song and the rest, and ordering
//! the songs by distance, ascending.
//!
//! If you want to implement a bliss plugin for an already existing audio
//! player, the [Library] struct is a collection of goodies that should prove
//! useful (it contains utilities to store analyzed songs in a self-contained
//! database file, to make playlists directly from the database, etc).
//! [blissify](https://github.com/Polochon-street/blissify-rs/) for both
//! an example of how the [Library] struct works, and a real-life demo of bliss
//! implemented for [MPD](https://www.musicpd.org/).
//!
//! # Examples
//!
//! ### Analyze & compute the distance between two songs
@ -37,40 +29,9 @@
//! Ok(())
//! }
//! ```
//!
//! ### Make a playlist from a song, discarding failed songs
//! ```no_run
//! use bliss_audio::{
//! analyze_paths,
//! playlist::{closest_to_first_song, euclidean_distance},
//! BlissResult, Song,
//! };
//!
//! fn main() -> BlissResult<()> {
//! let paths = vec!["/path/to/song1", "/path/to/song2", "/path/to/song3"];
//! let mut songs: Vec<Song> = analyze_paths(&paths).filter_map(|(_, s)| s.ok()).collect();
//!
//! // Assuming there is a first song
//! let first_song = songs.first().unwrap().to_owned();
//!
//! closest_to_first_song(&first_song, &mut songs, euclidean_distance);
//!
//! println!("Playlist is:");
//! for song in songs {
//! println!("{}", song.path.display());
//! }
//! Ok(())
//! }
//! ```
#![cfg_attr(feature = "bench", feature(test))]
#![warn(missing_docs)]
use crate::cue::BlissCue;
use log::info;
use std::num::NonZeroUsize;
use std::path::{Path, PathBuf};
use std::sync::mpsc;
use std::thread;
use thiserror::Error;
pub use crate::song::{Analysis, AnalysisIndex, Song, NUMBER_FEATURES};
@ -103,147 +64,9 @@ pub enum BlissError {
/// bliss error type
pub type BlissResult<T> = Result<T, BlissError>;
/// Analyze songs in `paths`, and return the analyzed [Song] objects through an
/// [mpsc::IntoIter].
///
/// Returns an iterator, whose items are a tuple made of
/// the song path (to display to the user in case the analysis failed),
/// and a Result<Song>.
///
/// # Note
///
/// This function also works with CUE files - it finds the audio files
/// mentionned in the CUE sheet, and then runs the analysis on each song
/// defined by it, returning a proper [Song] object for each one of them.
///
/// Make sure that you don't submit both the audio file along with the CUE
/// sheet if your library uses them, otherwise the audio file will be
/// analyzed as one, single, long song. For instance, with a CUE sheet named
/// `cue-file.cue` with the corresponding audio files `album-1.wav` and
/// `album-2.wav` defined in the CUE sheet, you would just pass `cue-file.cue`
/// to `analyze_paths`, and it will return [Song]s from both files, with
/// more information about which file it is extracted from in the
/// [cue info field](Song::cue_info).
///
/// # Example:
/// ```no_run
/// use bliss_audio::{analyze_paths, BlissResult};
///
/// fn main() -> BlissResult<()> {
/// let paths = vec![String::from("/path/to/song1"), String::from("/path/to/song2")];
/// for (path, result) in analyze_paths(&paths) {
/// match result {
/// Ok(song) => println!("Do something with analyzed song {} with title {:?}", song.path.display(), song.title),
/// Err(e) => println!("Song at {} could not be analyzed. Failed with: {}", path.display(), e),
/// }
/// }
/// Ok(())
/// }
/// ```
pub fn analyze_paths<P: Into<PathBuf>, F: IntoIterator<Item = P>>(
paths: F,
) -> mpsc::IntoIter<(PathBuf, BlissResult<Song>)> {
let cores = NonZeroUsize::new(num_cpus::get()).unwrap();
analyze_paths_with_cores(paths, cores)
}
/// Analyze songs in `paths`, and return the analyzed [Song] objects through an
/// [mpsc::IntoIter]. `number_cores` sets the number of cores the analysis
/// will use, capped by your system's capacity. Most of the time, you want to
/// use the simpler `analyze_paths` functions, which autodetects the number
/// of cores in your system.
///
/// Return an iterator, whose items are a tuple made of
/// the song path (to display to the user in case the analysis failed),
/// and a Result<Song>.
///
/// # Note
///
/// This function also works with CUE files - it finds the audio files
/// mentionned in the CUE sheet, and then runs the analysis on each song
/// defined by it, returning a proper [Song] object for each one of them.
///
/// Make sure that you don't submit both the audio file along with the CUE
/// sheet if your library uses them, otherwise the audio file will be
/// analyzed as one, single, long song. For instance, with a CUE sheet named
/// `cue-file.cue` with the corresponding audio files `album-1.wav` and
/// `album-2.wav` defined in the CUE sheet, you would just pass `cue-file.cue`
/// to `analyze_paths`, and it will return [Song]s from both files, with
/// more information about which file it is extracted from in the
/// [cue info field](Song::cue_info).
///
/// # Example:
/// ```no_run
/// use bliss_audio::{analyze_paths, BlissResult};
///
/// fn main() -> BlissResult<()> {
/// let paths = vec![String::from("/path/to/song1"), String::from("/path/to/song2")];
/// for (path, result) in analyze_paths(&paths) {
/// match result {
/// Ok(song) => println!("Do something with analyzed song {} with title {:?}", song.path.display(), song.title),
/// Err(e) => println!("Song at {} could not be analyzed. Failed with: {}", path.display(), e),
/// }
/// }
/// Ok(())
/// }
/// ```
pub fn analyze_paths_with_cores<P: Into<PathBuf>, F: IntoIterator<Item = P>>(
paths: F,
number_cores: NonZeroUsize,
) -> mpsc::IntoIter<(PathBuf, BlissResult<Song>)> {
let mut cores = NonZeroUsize::new(num_cpus::get()).unwrap();
if cores > number_cores {
cores = number_cores;
}
let paths: Vec<PathBuf> = paths.into_iter().map(|p| p.into()).collect();
#[allow(clippy::type_complexity)]
let (tx, rx): (
mpsc::Sender<(PathBuf, BlissResult<Song>)>,
mpsc::Receiver<(PathBuf, BlissResult<Song>)>,
) = mpsc::channel();
if paths.is_empty() {
return rx.into_iter();
}
let mut handles = Vec::new();
let mut chunk_length = paths.len() / cores;
if chunk_length == 0 {
chunk_length = paths.len();
}
for chunk in paths.chunks(chunk_length) {
let tx_thread = tx.clone();
let owned_chunk = chunk.to_owned();
let child = thread::spawn(move || {
for path in owned_chunk {
info!("Analyzing file '{:?}'", path);
if let Some(extension) = Path::new(&path).extension() {
let extension = extension.to_string_lossy().to_lowercase();
if extension == "cue" {
match BlissCue::songs_from_path(&path) {
Ok(songs) => {
for song in songs {
tx_thread.send((path.to_owned(), song)).unwrap();
}
}
Err(e) => tx_thread.send((path.to_owned(), Err(e))).unwrap(),
};
continue;
}
}
let song = Song::from_path(&path);
tx_thread.send((path.to_owned(), song)).unwrap();
}
});
handles.push(child);
}
rx.into_iter()
}
#[cfg(test)]
mod tests {
use super::*;
#[cfg(test)]
use pretty_assertions::assert_eq;
#[test]
fn test_send_song() {
@ -256,67 +79,4 @@ mod tests {
fn assert_sync<T: Send>() {}
assert_sync::<Song>();
}
#[test]
fn test_analyze_paths() {
let paths = vec![
"./data/s16_mono_22_5kHz.flac",
"./data/testcue.cue",
"./data/white_noise.flac",
"definitely-not-existing.foo",
"not-existing.foo",
];
let mut results = analyze_paths(&paths)
.map(|x| match &x.1 {
Ok(s) => (true, s.path.to_owned(), None),
Err(e) => (false, x.0.to_owned(), Some(e.to_string())),
})
.collect::<Vec<_>>();
results.sort();
let expected_results = vec![
(
false,
PathBuf::from("./data/testcue.cue"),
Some(String::from(
"error happened while decoding file – while \
opening format for file './data/not-existing.wav': \
ffmpeg::Error(2: No such file or directory).",
)),
),
(
false,
PathBuf::from("definitely-not-existing.foo"),
Some(String::from(
"error happened while decoding file – while \
opening format for file 'definitely-not-existing\
.foo': ffmpeg::Error(2: No such file or directory).",
)),
),
(
false,
PathBuf::from("not-existing.foo"),
Some(String::from(
"error happened while decoding file – \
while opening format for file 'not-existing.foo': \
ffmpeg::Error(2: No such file or directory).",
)),
),
(true, PathBuf::from("./data/s16_mono_22_5kHz.flac"), None),
(true, PathBuf::from("./data/testcue.cue/CUE_TRACK001"), None),
(true, PathBuf::from("./data/testcue.cue/CUE_TRACK002"), None),
(true, PathBuf::from("./data/testcue.cue/CUE_TRACK003"), None),
(true, PathBuf::from("./data/white_noise.flac"), None),
];
assert_eq!(results, expected_results);
let mut results = analyze_paths_with_cores(&paths, NonZeroUsize::new(1).unwrap())
.map(|x| match &x.1 {
Ok(s) => (true, s.path.to_owned(), None),
Err(e) => (false, x.0.to_owned(), Some(e.to_string())),
})
.collect::<Vec<_>>();
results.sort();
assert_eq!(results, expected_results);
}
}

339
src/cue.rs
View File

@ -1,339 +0,0 @@
//! CUE-handling module.
//!
//! Using [BlissCue::songs_from_path] is most likely what you want.
use crate::bliss_lib::{Analysis, BlissError, BlissResult, Song, FEATURES_VERSION, SAMPLE_RATE};
use rcue::cue::{Cue, Track};
use rcue::parser::parse_from_file;
use std::path::{Path, PathBuf};
use std::time::Duration;
#[cfg_attr(feature = "serde", derive(Serialize, Deserialize))]
#[derive(Default, Debug, PartialEq, Eq, Clone)]
/// A struct populated when the corresponding [Song] has been extracted from an
/// audio file split with the help of a CUE sheet.
pub struct CueInfo {
/// The path of the original CUE sheet, e.g. `/path/to/album_name.cue`.
pub cue_path: PathBuf,
/// The path of the audio file the song was extracted from, e.g.
/// `/path/to/album_name.wav`. Used because one CUE sheet can refer to
/// several audio files.
pub audio_file_path: PathBuf,
}
/// A struct to handle CUEs with bliss.
/// Use either [analyze_paths](crate::analyze_paths) with CUE files or
/// [songs_from_path](BlissCue::songs_from_path) to return a list of [Song]s
/// from CUE files.
pub struct BlissCue {
cue: Cue,
cue_path: PathBuf,
}
#[allow(missing_docs)]
#[derive(Default, Debug, PartialEq, Clone)]
struct BlissCueFile {
sample_array: Vec<f32>,
album: Option<String>,
artist: Option<String>,
genre: Option<String>,
tracks: Vec<Track>,
cue_path: PathBuf,
audio_file_path: PathBuf,
}
impl BlissCue {
/// Analyze songs from a CUE file, extracting individual [Song] objects
/// for each individual song.
///
/// Each returned [Song] has a populated [cue_info](Song::cue_info) object, that can be
/// be used to retrieve which CUE sheet was used to extract it, as well
/// as the corresponding audio file.
pub fn songs_from_path<P: AsRef<Path>>(path: P) -> BlissResult<Vec<BlissResult<Song>>> {
let cue = BlissCue::from_path(&path)?;
let cue_files = cue.files();
let mut songs = Vec::new();
for cue_file in cue_files.into_iter() {
match cue_file {
Ok(f) => {
if !f.sample_array.is_empty() {
songs.extend_from_slice(&f.get_songs());
} else {
songs.push(Err(BlissError::DecodingError(
"empty audio file associated to CUE sheet".into(),
)));
}
}
Err(e) => songs.push(Err(e)),
}
}
Ok(songs)
}
// Extract a BlissCue from a given path.
fn from_path<P: AsRef<Path>>(path: P) -> BlissResult<Self> {
let cue = parse_from_file(&path.as_ref().to_string_lossy(), false).map_err(|e| {
BlissError::DecodingError(format!(
"when opening CUE file '{:?}': {:?}",
path.as_ref(),
e
))
})?;
Ok(BlissCue {
cue,
cue_path: path.as_ref().to_owned(),
})
}
// List all BlissCueFile from a BlissCue.
fn files(&self) -> Vec<BlissResult<BlissCueFile>> {
let mut cue_files = Vec::new();
for cue_file in self.cue.files.iter() {
let audio_file_path = match &self.cue_path.parent() {
Some(parent) => parent.join(Path::new(&cue_file.file)),
None => PathBuf::from(cue_file.file.to_owned()),
};
let genre = self
.cue
.comments
.iter()
.find(|(c, _)| c == "GENRE")
.map(|(_, v)| v.to_owned());
let raw_song = Song::decode(Path::new(&audio_file_path));
if let Ok(song) = raw_song {
let bliss_cue_file = BlissCueFile {
sample_array: song.sample_array,
genre,
artist: self.cue.performer.to_owned(),
album: self.cue.title.to_owned(),
tracks: cue_file.tracks.to_owned(),
audio_file_path,
cue_path: self.cue_path.to_owned(),
};
cue_files.push(Ok(bliss_cue_file))
} else {
cue_files.push(Err(raw_song.unwrap_err()));
}
}
cue_files
}
}
impl BlissCueFile {
fn create_song(
&self,
analysis: BlissResult<Analysis>,
current_track: &Track,
duration: Duration,
index: usize,
) -> BlissResult<Song> {
if let Ok(a) = analysis {
let song = Song {
path: PathBuf::from(format!(
"{}/CUE_TRACK{:03}",
self.cue_path.to_string_lossy(),
index,
)),
album: self.album.to_owned(),
artist: current_track.performer.to_owned(),
album_artist: self.artist.to_owned(),
analysis: a,
duration,
genre: self.genre.to_owned(),
title: current_track.title.to_owned(),
track_number: Some(current_track.no.to_owned()),
features_version: FEATURES_VERSION,
cue_info: Some(CueInfo {
cue_path: self.cue_path.to_owned(),
audio_file_path: self.audio_file_path.to_owned(),
}),
};
Ok(song)
} else {
Err(analysis.unwrap_err())
}
}
// Get all songs from a BlissCueFile, using Song::analyze, each song being
// located using the sample_array and the timestamp delimiter.
fn get_songs(&self) -> Vec<BlissResult<Song>> {
let mut songs = Vec::new();
for (index, tuple) in (self.tracks[..]).windows(2).enumerate() {
let (current_track, next_track) = (tuple[0].to_owned(), tuple[1].to_owned());
if let Some((_, start_current)) = current_track.indices.get(0) {
if let Some((_, end_current)) = next_track.indices.get(0) {
let start_current = (start_current.as_secs_f32() * SAMPLE_RATE as f32) as usize;
let end_current = (end_current.as_secs_f32() * SAMPLE_RATE as f32) as usize;
let duration = Duration::from_secs_f32(
(end_current - start_current) as f32 / SAMPLE_RATE as f32,
);
let analysis = Song::analyze(&self.sample_array[start_current..end_current]);
let song = self.create_song(analysis, &current_track, duration, index + 1);
songs.push(song);
}
}
}
// Take care of the last track, since the windows iterator doesn't.
if let Some(last_track) = self.tracks.last() {
if let Some((_, start_current)) = last_track.indices.get(0) {
let start_current = (start_current.as_secs_f32() * SAMPLE_RATE as f32) as usize;
let duration = Duration::from_secs_f32(
(self.sample_array.len() - start_current) as f32 / SAMPLE_RATE as f32,
);
let analysis = Song::analyze(&self.sample_array[start_current..]);
let song = self.create_song(analysis, last_track, duration, self.tracks.len());
songs.push(song);
}
}
songs
}
}
#[cfg(test)]
mod tests {
use super::*;
use pretty_assertions::assert_eq;
#[test]
fn test_empty_cue() {
let songs = BlissCue::songs_from_path("data/empty.cue").unwrap();
let error = songs[0].to_owned().unwrap_err();
assert_eq!(
error,
BlissError::DecodingError("empty audio file associated to CUE sheet".to_string())
);
}
#[test]
fn test_cue_analysis() {
let songs = BlissCue::songs_from_path("data/testcue.cue").unwrap();
let expected = vec![
Ok(Song {
path: Path::new("data/testcue.cue/CUE_TRACK001").to_path_buf(),
analysis: Analysis {
internal_analysis: [
0.38463724,
-0.85219246,
-0.761946,
-0.8904667,
-0.63892543,
-0.73945934,
-0.8004017,
-0.8237293,
0.33865356,
0.32481194,
-0.35692245,
-0.6355889,
-0.29584837,
0.06431806,
0.21875131,
-0.58104205,
-0.9466792,
-0.94811195,
-0.9820919,
-0.9596871,
],
},
album: Some(String::from("Album for CUE test")),
artist: Some(String::from("David TMX")),
title: Some(String::from("Renaissance")),
genre: Some(String::from("Random")),
track_number: Some(String::from("01")),
features_version: FEATURES_VERSION,
album_artist: Some(String::from("Polochon_street")),
duration: Duration::from_secs_f32(11.066666603),
cue_info: Some(CueInfo {
cue_path: PathBuf::from("data/testcue.cue"),
audio_file_path: PathBuf::from("data/testcue.flac"),
}),
..Default::default()
}),
Ok(Song {
path: Path::new("data/testcue.cue/CUE_TRACK002").to_path_buf(),
analysis: Analysis {
internal_analysis: [
0.18622077,
-0.5989029,
-0.5554645,
-0.6343865,
-0.24163479,
-0.25766593,
-0.40616858,
-0.23334873,
0.76875293,
0.7785741,
-0.5075115,
-0.5272629,
-0.56706166,
-0.568486,
-0.5639081,
-0.5706943,
-0.96501005,
-0.96501285,
-0.9649896,
-0.96498996,
],
},
features_version: FEATURES_VERSION,
album: Some(String::from("Album for CUE test")),
artist: Some(String::from("Polochon_street")),
title: Some(String::from("Piano")),
genre: Some(String::from("Random")),
track_number: Some(String::from("02")),
album_artist: Some(String::from("Polochon_street")),
duration: Duration::from_secs_f64(5.853333473),
cue_info: Some(CueInfo {
cue_path: PathBuf::from("data/testcue.cue"),
audio_file_path: PathBuf::from("data/testcue.flac"),
}),
..Default::default()
}),
Ok(Song {
path: Path::new("data/testcue.cue/CUE_TRACK003").to_path_buf(),
analysis: Analysis {
internal_analysis: [
0.0024261475,
0.9874661,
0.97330654,
-0.9724426,
0.99678576,
-0.9961549,
-0.9840142,
-0.9269961,
0.7498772,
0.22429907,
-0.8355152,
-0.9977258,
-0.9977849,
-0.997785,
-0.99778515,
-0.997785,
-0.99999976,
-0.99999976,
-0.99999976,
-0.99999976,
],
},
album: Some(String::from("Album for CUE test")),
artist: Some(String::from("Polochon_street")),
title: Some(String::from("Tone")),
genre: Some(String::from("Random")),
track_number: Some(String::from("03")),
features_version: FEATURES_VERSION,
album_artist: Some(String::from("Polochon_street")),
duration: Duration::from_secs_f32(5.586666584),
cue_info: Some(CueInfo {
cue_path: PathBuf::from("data/testcue.cue"),
audio_file_path: PathBuf::from("data/testcue.flac"),
}),
..Default::default()
}),
Err(BlissError::DecodingError(String::from(
"while opening format for file 'data/not-existing.wav': \
ffmpeg::Error(2: No such file or directory).",
))),
];
assert_eq!(expected, songs);
}
}

30
src/lib.rs
View File

@ -1,24 +1,14 @@
extern crate crossbeam;
extern crate num_cpus;
#[cfg(feature = "serde")]
#[macro_use]
extern crate serde;
pub mod bliss_lib;
mod chroma;
pub mod cue;
#[cfg(feature = "library")]
pub mod library;
mod misc;
pub mod playlist;
mod song;
mod misc;
mod temporal;
mod timbral;
mod utils;
use neon::{prelude::*, types::buffer::TypedArray};
use song::Song;
use bliss_lib::BlissResult;
// use song::Song;
// use bliss_lib::BlissResult;
#[neon::main]
fn main(mut cx: ModuleContext) -> NeonResult<()> {
@ -69,9 +59,13 @@ fn analyze(mut cx: FunctionContext) -> JsResult<JsUint8Array> {
Ok(buffer_handle)
}
fn analyze_raw(path: &str) -> BlissResult<([u8; 2], [u8; 80])> {
let song = Song::from_path(path)?;
let version_bytes = song.features_version.to_le_bytes();
let analysis_bytes = song.analysis.as_bytes();
Ok((version_bytes, analysis_bytes))
// fn analyze_raw(path: &str) -> BlissResult<([u8; 2], [u8; 80])> {
// let song = Song::from_path(path)?;
// let version_bytes = song.features_version.to_le_bytes();
// let analysis_bytes = song.analysis.as_bytes();
// Ok((version_bytes, analysis_bytes))
// }
fn analyze_raw(path: &str) -> Result<([u8; 2], [u8; 80]), u8> {
return Ok(([0; 2], [0; 80]));
}

3183
src/library.rs
View File

File diff suppressed because it is too large Load Diff

984
src/playlist.rs
View File

@ -1,984 +0,0 @@
//! Module containing various functions to build playlists, as well as various
//! distance metrics.
//!
//! All of the distance functions are intended to be used with the
//! [custom_distance](Song::custom_distance) method, or with
//!
//! They will yield different styles of playlists, so don't hesitate to
//! experiment with them if the default (euclidean distance for now) doesn't
//! suit you.
// TODO on the `by_key` functions: maybe Fn(&T) -> &Song is enough? Compared
// to -> Song
use crate::bliss_lib::{BlissError, BlissResult, Song, NUMBER_FEATURES};
use ndarray::{Array, Array1, Array2, Axis};
use ndarray_stats::QuantileExt;
use noisy_float::prelude::*;
use std::collections::HashMap;
/// Convenience trait for user-defined distance metrics.
pub trait DistanceMetric: Fn(&Array1<f32>, &Array1<f32>) -> f32 {}
impl<F> DistanceMetric for F where F: Fn(&Array1<f32>, &Array1<f32>) -> f32 {}
/// Return the [euclidean
/// distance](https://en.wikipedia.org/wiki/Euclidean_distance#Higher_dimensions)
/// between two vectors.
pub fn euclidean_distance(a: &Array1<f32>, b: &Array1<f32>) -> f32 {
// Could be any square symmetric positive semi-definite matrix;
// just no metric learning has been done yet.
// See https://lelele.io/thesis.pdf chapter 4.
let m = Array::eye(NUMBER_FEATURES);
(a - b).dot(&m).dot(&(a - b)).sqrt()
}
/// Return the [cosine
/// distance](https://en.wikipedia.org/wiki/Cosine_similarity#Angular_distance_and_similarity)
/// between two vectors.
pub fn cosine_distance(a: &Array1<f32>, b: &Array1<f32>) -> f32 {
let similarity = a.dot(b) / (a.dot(a).sqrt() * b.dot(b).sqrt());
1. - similarity
}
/// Sort `songs` in place by putting songs close to `first_song` first
/// using the `distance` metric.
pub fn closest_to_first_song(
first_song: &Song,
#[allow(clippy::ptr_arg)] songs: &mut Vec<Song>,
distance: impl DistanceMetric,
) {
songs.sort_by_cached_key(|song| n32(first_song.custom_distance(song, &distance)));
}
/// Sort `songs` in place by putting songs close to `first_song` first
/// using the `distance` metric.
///
/// Sort songs with a key extraction function, useful for when you have a
/// structure like `CustomSong { bliss_song: Song, something_else: bool }`
pub fn closest_to_first_song_by_key<F, T>(
first_song: &T,
#[allow(clippy::ptr_arg)] songs: &mut Vec<T>,
distance: impl DistanceMetric,
key_fn: F,
) where
F: Fn(&T) -> Song,
{
let first_song = key_fn(first_song);
songs.sort_by_cached_key(|song| n32(first_song.custom_distance(&key_fn(song), &distance)));
}
/// Sort `songs` in place using the `distance` metric and ordering by
/// the smallest distance between each song.
///
/// If the generated playlist is `[song1, song2, song3, song4]`, it means
/// song2 is closest to song1, song3 is closest to song2, and song4 is closest
/// to song3.
///
/// Note that this has a tendency to go from one style to the other very fast,
/// and it can be slow on big libraries.
pub fn song_to_song(first_song: &Song, songs: &mut Vec<Song>, distance: impl DistanceMetric) {
let mut new_songs = Vec::with_capacity(songs.len());
let mut song = first_song.to_owned();
while !songs.is_empty() {
let distances: Array1<f32> =
Array::from_shape_fn(songs.len(), |i| song.custom_distance(&songs[i], &distance));
let idx = distances.argmin().unwrap();
song = songs[idx].to_owned();
new_songs.push(song.to_owned());
songs.retain(|s| s != &song);
}
*songs = new_songs;
}
/// Sort `songs` in place using the `distance` metric and ordering by
/// the smallest distance between each song.
///
/// If the generated playlist is `[song1, song2, song3, song4]`, it means
/// song2 is closest to song1, song3 is closest to song2, and song4 is closest
/// to song3.
///
/// Note that this has a tendency to go from one style to the other very fast,
/// and it can be slow on big libraries.
///
/// Sort songs with a key extraction function, useful for when you have a
/// structure like `CustomSong { bliss_song: Song, something_else: bool }`
// TODO: maybe Clone is not needed?
pub fn song_to_song_by_key<F, T: std::cmp::PartialEq + Clone>(
first_song: &T,
songs: &mut Vec<T>,
distance: impl DistanceMetric,
key_fn: F,
) where
F: Fn(&T) -> Song,
{
let mut new_songs: Vec<T> = Vec::with_capacity(songs.len());
let mut bliss_song = key_fn(&first_song.to_owned());
while !songs.is_empty() {
let distances: Array1<f32> = Array::from_shape_fn(songs.len(), |i| {
bliss_song.custom_distance(&key_fn(&songs[i]), &distance)
});
let idx = distances.argmin().unwrap();
let song = songs[idx].to_owned();
bliss_song = key_fn(&songs[idx]).to_owned();
new_songs.push(song.to_owned());
songs.retain(|s| s != &song);
}
*songs = new_songs;
}
/// Remove duplicate songs from a playlist, in place.
///
/// Two songs are considered duplicates if they either have the same,
/// non-empty title and artist name, or if they are close enough in terms
/// of distance.
///
/// # Arguments
///
/// * `songs`: The playlist to remove duplicates from.
/// * `distance_threshold`: The distance threshold under which two songs are
/// considered identical. If `None`, a default value of 0.05 will be used.
pub fn dedup_playlist(songs: &mut Vec<Song>, distance_threshold: Option<f32>) {
dedup_playlist_custom_distance(songs, distance_threshold, euclidean_distance);
}
/// Remove duplicate songs from a playlist, in place.
///
/// Two songs are considered duplicates if they either have the same,
/// non-empty title and artist name, or if they are close enough in terms
/// of distance.
///
/// Dedup songs with a key extraction function, useful for when you have a
/// structure like `CustomSong { bliss_song: Song, something_else: bool }` you
/// want to deduplicate.
///
/// # Arguments
///
/// * `songs`: The playlist to remove duplicates from.
/// * `distance_threshold`: The distance threshold under which two songs are
/// considered identical. If `None`, a default value of 0.05 will be used.
/// * `key_fn`: A function used to retrieve the bliss [Song] from `T`.
pub fn dedup_playlist_by_key<T, F>(songs: &mut Vec<T>, distance_threshold: Option<f32>, key_fn: F)
where
F: Fn(&T) -> Song,
{
dedup_playlist_custom_distance_by_key(songs, distance_threshold, euclidean_distance, key_fn);
}
/// Remove duplicate songs from a playlist, in place, using a custom distance
/// metric.
///
/// Two songs are considered duplicates if they either have the same,
/// non-empty title and artist name, or if they are close enough in terms
/// of distance.
///
/// # Arguments
///
/// * `songs`: The playlist to remove duplicates from.
/// * `distance_threshold`: The distance threshold under which two songs are
/// considered identical. If `None`, a default value of 0.05 will be used.
/// * `distance`: A custom distance metric.
pub fn dedup_playlist_custom_distance(
songs: &mut Vec<Song>,
distance_threshold: Option<f32>,
distance: impl DistanceMetric,
) {
songs.dedup_by(|s1, s2| {
n32(s1.custom_distance(s2, &distance)) < distance_threshold.unwrap_or(0.05)
|| (s1.title.is_some()
&& s2.title.is_some()
&& s1.artist.is_some()
&& s2.artist.is_some()
&& s1.title == s2.title
&& s1.artist == s2.artist)
});
}
/// Remove duplicate songs from a playlist, in place, using a custom distance
/// metric.
///
/// Two songs are considered duplicates if they either have the same,
/// non-empty title and artist name, or if they are close enough in terms
/// of distance.
///
/// Dedup songs with a key extraction function, useful for when you have a
/// structure like `CustomSong { bliss_song: Song, something_else: bool }`
/// you want to deduplicate.
///
/// # Arguments
///
/// * `songs`: The playlist to remove duplicates from.
/// * `distance_threshold`: The distance threshold under which two songs are
/// considered identical. If `None`, a default value of 0.05 will be used.
/// * `distance`: A custom distance metric.
/// * `key_fn`: A function used to retrieve the bliss [Song] from `T`.
pub fn dedup_playlist_custom_distance_by_key<F, T>(
songs: &mut Vec<T>,
distance_threshold: Option<f32>,
distance: impl DistanceMetric,
key_fn: F,
) where
F: Fn(&T) -> Song,
{
songs.dedup_by(|s1, s2| {
let s1 = key_fn(s1);
let s2 = key_fn(s2);
n32(s1.custom_distance(&s2, &distance)) < distance_threshold.unwrap_or(0.05)
|| (s1.title.is_some()
&& s2.title.is_some()
&& s1.artist.is_some()
&& s2.artist.is_some()
&& s1.title == s2.title
&& s1.artist == s2.artist)
});
}
/// Return a list of albums in a `pool` of songs that are similar to
/// songs in `group`, discarding songs that don't belong to an album.
/// It basically makes an "album" playlist from the `pool` of songs.
///
/// `group` should be ordered by track number.
///
/// Songs from `group` would usually just be songs from an album, but not
/// necessarily - they are discarded from `pool` no matter what.
///
/// # Arguments
///
/// * `group` - A small group of songs, e.g. an album.
/// * `pool` - A pool of songs to find similar songs in, e.g. a user's song
/// library.
///
/// # Returns
///
/// A vector of songs, including `group` at the beginning, that you
/// most likely want to plug in your audio player by using something like
/// `ret.map(|song| song.path.to_owned()).collect::<Vec<String>>()`.
pub fn closest_album_to_group(group: Vec<Song>, pool: Vec<Song>) -> BlissResult<Vec<Song>> {
let mut albums_analysis: HashMap<&str, Array2<f32>> = HashMap::new();
let mut albums = Vec::new();
// Remove songs from the group from the pool.
let pool = pool
.into_iter()
.filter(|s| !group.contains(s))
.collect::<Vec<_>>();
for song in &pool {
if let Some(album) = &song.album {
if let Some(analysis) = albums_analysis.get_mut(album as &str) {
analysis
.push_row(song.analysis.as_arr1().view())
.map_err(|e| {
BlissError::ProviderError(format!("while computing distances: {e}"))
})?;
} else {
let mut array = Array::zeros((1, song.analysis.as_arr1().len()));
array.assign(&song.analysis.as_arr1());
albums_analysis.insert(album, array);
}
}
}
let mut group_analysis = Array::zeros((group.len(), NUMBER_FEATURES));
for (song, mut column) in group.iter().zip(group_analysis.axis_iter_mut(Axis(0))) {
column.assign(&song.analysis.as_arr1());
}
let first_analysis = group_analysis
.mean_axis(Axis(0))
.ok_or_else(|| BlissError::ProviderError(String::from("Mean of empty slice")))?;
for (album, analysis) in albums_analysis.iter() {
let mean_analysis = analysis
.mean_axis(Axis(0))
.ok_or_else(|| BlissError::ProviderError(String::from("Mean of empty slice")))?;
let album = album.to_owned();
albums.push((album, mean_analysis.to_owned()));
}
albums.sort_by_key(|(_, analysis)| n32(euclidean_distance(&first_analysis, analysis)));
let mut playlist = group;
for (album, _) in albums {
let mut al = pool
.iter()
.filter(|s| s.album.is_some() && s.album.as_ref().unwrap() == &album.to_string())
.map(|s| s.to_owned())
.collect::<Vec<Song>>();
al.sort_by(|s1, s2| {
let track_number1 = s1
.track_number
.to_owned()
.unwrap_or_else(|| String::from(""));
let track_number2 = s2
.track_number
.to_owned()
.unwrap_or_else(|| String::from(""));
if let Ok(x) = track_number1.parse::<i32>() {
if let Ok(y) = track_number2.parse::<i32>() {
return x.cmp(&y);
}
}
s1.track_number.cmp(&s2.track_number)
});
playlist.extend_from_slice(&al);
}
Ok(playlist)
}
/// Return a list of albums in a `pool` of songs that are similar to
/// songs in `group`, discarding songs that don't belong to an album.
/// It basically makes an "album" playlist from the `pool` of songs.
///
/// `group` should be ordered by track number.
///
/// Songs from `group` would usually just be songs from an album, but not
/// necessarily - they are discarded from `pool` no matter what.
///
/// Order songs with a key extraction function, useful for when you have a
/// structure like `CustomSong { bliss_song: Song, something_else: bool }`
/// you want to order.
///
/// # Arguments
///
/// * `group` - A small group of songs, e.g. an album.
/// * `pool` - A pool of songs to find similar songs in, e.g. a user's song
/// library.
/// * `key_fn`: A function used to retrieve the bliss [Song] from `T`.
///
/// # Returns
///
/// A vector of T, including `group` at the beginning, that you
/// most likely want to plug in your audio player by using something like
/// `ret.map(|song| song.path.to_owned()).collect::<Vec<String>>()`.
// TODO: maybe Clone is not needed?
pub fn closest_album_to_group_by_key<T: PartialEq + Clone, F>(
group: Vec<T>,
pool: Vec<T>,
key_fn: F,
) -> BlissResult<Vec<T>>
where
F: Fn(&T) -> Song,
{
let mut albums_analysis: HashMap<String, Array2<f32>> = HashMap::new();
let mut albums = Vec::new();
// Remove songs from the group from the pool.
let pool = pool
.into_iter()
.filter(|s| !group.contains(s))
.collect::<Vec<_>>();
for song in &pool {
let song = key_fn(song);
if let Some(album) = song.album {
if let Some(analysis) = albums_analysis.get_mut(&album as &str) {
analysis
.push_row(song.analysis.as_arr1().view())
.map_err(|e| {
BlissError::ProviderError(format!("while computing distances: {e}"))
})?;
} else {
let mut array = Array::zeros((1, song.analysis.as_arr1().len()));
array.assign(&song.analysis.as_arr1());
albums_analysis.insert(album.to_owned(), array);
}
}
}
let mut group_analysis = Array::zeros((group.len(), NUMBER_FEATURES));
for (song, mut column) in group.iter().zip(group_analysis.axis_iter_mut(Axis(0))) {
let song = key_fn(song);
column.assign(&song.analysis.as_arr1());
}
let first_analysis = group_analysis
.mean_axis(Axis(0))
.ok_or_else(|| BlissError::ProviderError(String::from("Mean of empty slice")))?;
for (album, analysis) in albums_analysis.iter() {
let mean_analysis = analysis
.mean_axis(Axis(0))
.ok_or_else(|| BlissError::ProviderError(String::from("Mean of empty slice")))?;
let album = album.to_owned();
albums.push((album, mean_analysis.to_owned()));
}
albums.sort_by_key(|(_, analysis)| n32(euclidean_distance(&first_analysis, analysis)));
let mut playlist = group;
for (album, _) in albums {
let mut al = pool
.iter()
.filter(|s| {
let s = key_fn(s);
s.album.is_some() && s.album.as_ref().unwrap() == &album.to_string()
})
.map(|s| s.to_owned())
.collect::<Vec<T>>();
al.sort_by(|s1, s2| {
let s1 = key_fn(s1);
let s2 = key_fn(s2);
let track_number1 = s1
.track_number
.to_owned()
.unwrap_or_else(|| String::from(""));
let track_number2 = s2
.track_number
.to_owned()
.unwrap_or_else(|| String::from(""));
if let Ok(x) = track_number1.parse::<i32>() {
if let Ok(y) = track_number2.parse::<i32>() {
return x.cmp(&y);
}
}
s1.track_number.cmp(&s2.track_number)
});
playlist.extend_from_slice(&al);
}
Ok(playlist)
}
#[cfg(test)]
mod test {
use super::*;
use crate::bliss_lib::Analysis;
use ndarray::arr1;
use std::path::Path;
#[derive(Debug, Clone, PartialEq)]
struct CustomSong {
something: bool,
bliss_song: Song,
}
#[test]
fn test_dedup_playlist_custom_distance() {
let first_song = Song {
path: Path::new("path-to-first").to_path_buf(),
analysis: Analysis::new([
1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1.,
]),
..Default::default()
};
let first_song_dupe = Song {
path: Path::new("path-to-dupe").to_path_buf(),
analysis: Analysis::new([
1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1.,
]),
..Default::default()
};
let second_song = Song {
path: Path::new("path-to-second").to_path_buf(),
analysis: Analysis::new([
2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 1.9, 1., 1., 1.,
]),
title: Some(String::from("dupe-title")),
artist: Some(String::from("dupe-artist")),
..Default::default()
};
let third_song = Song {
path: Path::new("path-to-third").to_path_buf(),
title: Some(String::from("dupe-title")),
artist: Some(String::from("dupe-artist")),
analysis: Analysis::new([
2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2.5, 1., 1., 1.,
]),
..Default::default()
};
let fourth_song = Song {
path: Path::new("path-to-fourth").to_path_buf(),
artist: Some(String::from("no-dupe-artist")),
title: Some(String::from("dupe-title")),
analysis: Analysis::new([
2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 0., 1., 1., 1.,
]),
..Default::default()
};
let fifth_song = Song {
path: Path::new("path-to-fourth").to_path_buf(),
analysis: Analysis::new([
2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 0.001, 1., 1., 1.,
]),
..Default::default()
};
let mut playlist = vec![
first_song.to_owned(),
first_song_dupe.to_owned(),
second_song.to_owned(),
third_song.to_owned(),
fourth_song.to_owned(),
fifth_song.to_owned(),
];
dedup_playlist_custom_distance(&mut playlist, None, euclidean_distance);
assert_eq!(
playlist,
vec![
first_song.to_owned(),
second_song.to_owned(),
fourth_song.to_owned(),
],
);
let mut playlist = vec![
first_song.to_owned(),
first_song_dupe.to_owned(),
second_song.to_owned(),
third_song.to_owned(),
fourth_song.to_owned(),
fifth_song.to_owned(),
];
dedup_playlist_custom_distance(&mut playlist, Some(20.), cosine_distance);
assert_eq!(playlist, vec![first_song.to_owned()]);
let mut playlist = vec![
first_song.to_owned(),
first_song_dupe.to_owned(),
second_song.to_owned(),
third_song.to_owned(),
fourth_song.to_owned(),
fifth_song.to_owned(),
];
dedup_playlist(&mut playlist, Some(20.));
assert_eq!(playlist, vec![first_song.to_owned()]);
let mut playlist = vec![
first_song.to_owned(),
first_song_dupe.to_owned(),
second_song.to_owned(),
third_song.to_owned(),
fourth_song.to_owned(),
fifth_song.to_owned(),
];
dedup_playlist(&mut playlist, None);
assert_eq!(
playlist,
vec![
first_song.to_owned(),
second_song.to_owned(),
fourth_song.to_owned(),
]
);
let first_song = CustomSong {
bliss_song: first_song,
something: true,
};
let second_song = CustomSong {
bliss_song: second_song,
something: true,
};
let first_song_dupe = CustomSong {
bliss_song: first_song_dupe,
something: true,
};
let third_song = CustomSong {
bliss_song: third_song,
something: true,
};
let fourth_song = CustomSong {
bliss_song: fourth_song,
something: true,
};
let fifth_song = CustomSong {
bliss_song: fifth_song,
something: true,
};
let mut playlist = vec![
first_song.to_owned(),
first_song_dupe.to_owned(),
second_song.to_owned(),
third_song.to_owned(),
fourth_song.to_owned(),
fifth_song.to_owned(),
];
dedup_playlist_custom_distance_by_key(&mut playlist, None, euclidean_distance, |s| {
s.bliss_song.to_owned()
});
assert_eq!(
playlist,
vec![
first_song.to_owned(),
second_song.to_owned(),
fourth_song.to_owned(),
],
);
let mut playlist = vec![
first_song.to_owned(),
first_song_dupe.to_owned(),
second_song.to_owned(),
third_song.to_owned(),
fourth_song.to_owned(),
fifth_song.to_owned(),
];
dedup_playlist_custom_distance_by_key(&mut playlist, Some(20.), cosine_distance, |s| {
s.bliss_song.to_owned()
});
assert_eq!(playlist, vec![first_song.to_owned()]);
let mut playlist = vec![
first_song.to_owned(),
first_song_dupe.to_owned(),
second_song.to_owned(),
third_song.to_owned(),
fourth_song.to_owned(),
fifth_song.to_owned(),
];
dedup_playlist_by_key(&mut playlist, Some(20.), |s| s.bliss_song.to_owned());
assert_eq!(playlist, vec![first_song.to_owned()]);
let mut playlist = vec![
first_song.to_owned(),
first_song_dupe.to_owned(),
second_song.to_owned(),
third_song.to_owned(),
fourth_song.to_owned(),
fifth_song.to_owned(),
];
dedup_playlist_by_key(&mut playlist, None, |s| s.bliss_song.to_owned());
assert_eq!(
playlist,
vec![
first_song.to_owned(),
second_song.to_owned(),
fourth_song.to_owned(),
]
);
}
#[test]
fn test_song_to_song() {
let first_song = Song {
path: Path::new("path-to-first").to_path_buf(),
analysis: Analysis::new([
1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1.,
]),
..Default::default()
};
let first_song_dupe = Song {
path: Path::new("path-to-dupe").to_path_buf(),
analysis: Analysis::new([
1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1.,
]),
..Default::default()
};
let second_song = Song {
path: Path::new("path-to-second").to_path_buf(),
analysis: Analysis::new([
2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 1.9, 1., 1., 1.,
]),
..Default::default()
};
let third_song = Song {
path: Path::new("path-to-third").to_path_buf(),
analysis: Analysis::new([
2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2.5, 1., 1., 1.,
]),
..Default::default()
};
let fourth_song = Song {
path: Path::new("path-to-fourth").to_path_buf(),
analysis: Analysis::new([
2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 0., 1., 1., 1.,
]),
..Default::default()
};
let mut songs = vec![
first_song.to_owned(),
third_song.to_owned(),
first_song_dupe.to_owned(),
second_song.to_owned(),
fourth_song.to_owned(),
];
song_to_song(&first_song, &mut songs, euclidean_distance);
assert_eq!(
songs,
vec![
first_song.to_owned(),
first_song_dupe.to_owned(),
second_song.to_owned(),
third_song.to_owned(),
fourth_song.to_owned(),
],
);
let first_song = CustomSong {
bliss_song: first_song,
something: true,
};
let second_song = CustomSong {
bliss_song: second_song,
something: true,
};
let first_song_dupe = CustomSong {
bliss_song: first_song_dupe,
something: true,
};
let third_song = CustomSong {
bliss_song: third_song,
something: true,
};
let fourth_song = CustomSong {
bliss_song: fourth_song,
something: true,
};
let mut songs: Vec<CustomSong> = vec![
first_song.to_owned(),
first_song_dupe.to_owned(),
third_song.to_owned(),
fourth_song.to_owned(),
second_song.to_owned(),
];
song_to_song_by_key(&first_song, &mut songs, euclidean_distance, |s| {
s.bliss_song.to_owned()
});
assert_eq!(
songs,
vec![
first_song,
first_song_dupe,
second_song,
third_song,
fourth_song,
],
);
}
#[test]
fn test_sort_closest_to_first_song() {
let first_song = Song {
path: Path::new("path-to-first").to_path_buf(),
analysis: Analysis::new([
1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1.,
]),
..Default::default()
};
let first_song_dupe = Song {
path: Path::new("path-to-dupe").to_path_buf(),
analysis: Analysis::new([
1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1.,
]),
..Default::default()
};
let second_song = Song {
path: Path::new("path-to-second").to_path_buf(),
analysis: Analysis::new([
2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 1.9, 1., 1., 1.,
]),
..Default::default()
};
let third_song = Song {
path: Path::new("path-to-third").to_path_buf(),
analysis: Analysis::new([
2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2.5, 1., 1., 1.,
]),
..Default::default()
};
let fourth_song = Song {
path: Path::new("path-to-fourth").to_path_buf(),
analysis: Analysis::new([
2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 0., 1., 1., 1.,
]),
..Default::default()
};
let fifth_song = Song {
path: Path::new("path-to-fifth").to_path_buf(),
analysis: Analysis::new([
2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 0., 1., 1., 1.,
]),
..Default::default()
};
let mut songs = vec![
first_song.to_owned(),
first_song_dupe.to_owned(),
second_song.to_owned(),
third_song.to_owned(),
fourth_song.to_owned(),
fifth_song.to_owned(),
];
closest_to_first_song(&first_song, &mut songs, euclidean_distance);
let first_song = CustomSong {
bliss_song: first_song,
something: true,
};
let second_song = CustomSong {
bliss_song: second_song,
something: true,
};
let first_song_dupe = CustomSong {
bliss_song: first_song_dupe,
something: true,
};
let third_song = CustomSong {
bliss_song: third_song,
something: true,
};
let fourth_song = CustomSong {
bliss_song: fourth_song,
something: true,
};
let fifth_song = CustomSong {
bliss_song: fifth_song,
something: true,
};
let mut songs: Vec<CustomSong> = vec![
first_song.to_owned(),
first_song_dupe.to_owned(),
second_song.to_owned(),
third_song.to_owned(),
fourth_song.to_owned(),
fifth_song.to_owned(),
];
closest_to_first_song_by_key(&first_song, &mut songs, euclidean_distance, |s| {
s.bliss_song.to_owned()
});
assert_eq!(
songs,
vec![
first_song,
first_song_dupe,
second_song,
fourth_song,
fifth_song,
third_song
],
);
}
#[test]
fn test_euclidean_distance() {
let a = arr1(&[
1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 0.,
]);
let b = arr1(&[
0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 1., 0., 0., 0.,
]);
assert_eq!(euclidean_distance(&a, &b), 4.242640687119285);
let a = arr1(&[0.5; 20]);
let b = arr1(&[0.5; 20]);
assert_eq!(euclidean_distance(&a, &b), 0.);
assert_eq!(euclidean_distance(&a, &b), 0.);
}
#[test]
fn test_cosine_distance() {
let a = arr1(&[
1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 0.,
]);
let b = arr1(&[
0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 1., 0., 0., 0.,
]);
assert_eq!(cosine_distance(&a, &b), 0.7705842661294382);
let a = arr1(&[0.5; 20]);
let b = arr1(&[0.5; 20]);
assert_eq!(cosine_distance(&a, &b), 0.);
assert_eq!(cosine_distance(&a, &b), 0.);
}
#[test]
fn test_closest_to_group() {
let first_song = Song {
path: Path::new("path-to-first").to_path_buf(),
analysis: Analysis::new([0.; 20]),
album: Some(String::from("Album")),
artist: Some(String::from("Artist")),
track_number: Some(String::from("01")),
..Default::default()
};
let second_song = Song {
path: Path::new("path-to-second").to_path_buf(),
analysis: Analysis::new([0.1; 20]),
album: Some(String::from("Another Album")),
artist: Some(String::from("Artist")),
track_number: Some(String::from("10")),
..Default::default()
};
let third_song = Song {
path: Path::new("path-to-third").to_path_buf(),
analysis: Analysis::new([10.; 20]),
album: Some(String::from("Album")),
artist: Some(String::from("Another Artist")),
track_number: Some(String::from("02")),
..Default::default()
};
let fourth_song = Song {
path: Path::new("path-to-fourth").to_path_buf(),
analysis: Analysis::new([20.; 20]),
album: Some(String::from("Another Album")),
artist: Some(String::from("Another Artist")),
track_number: Some(String::from("01")),
..Default::default()
};
let fifth_song = Song {
path: Path::new("path-to-fifth").to_path_buf(),
analysis: Analysis::new([40.; 20]),
artist: Some(String::from("Third Artist")),
album: None,
..Default::default()
};
let pool = vec![
first_song.to_owned(),
fourth_song.to_owned(),
third_song.to_owned(),
second_song.to_owned(),
fifth_song.to_owned(),
];
let group = vec![first_song.to_owned(), third_song.to_owned()];
assert_eq!(
vec![
first_song.to_owned(),
third_song.to_owned(),
fourth_song.to_owned(),
second_song.to_owned()
],
closest_album_to_group(group, pool.to_owned()).unwrap(),
);
let first_song = CustomSong {
bliss_song: first_song,
something: true,
};
let second_song = CustomSong {
bliss_song: second_song,
something: true,
};
let third_song = CustomSong {
bliss_song: third_song,
something: true,
};
let fourth_song = CustomSong {
bliss_song: fourth_song,
something: true,
};
let fifth_song = CustomSong {
bliss_song: fifth_song,
something: true,
};
let pool = vec![
first_song.to_owned(),
fourth_song.to_owned(),
third_song.to_owned(),
second_song.to_owned(),
fifth_song.to_owned(),
];
let group = vec![first_song.to_owned(), third_song.to_owned()];
assert_eq!(
vec![
first_song.to_owned(),
third_song.to_owned(),
fourth_song.to_owned(),
second_song.to_owned()
],
closest_album_to_group_by_key(group, pool.to_owned(), |s| s.bliss_song.to_owned())
.unwrap(),
);
}
}

396
src/song.rs
View File

@ -12,11 +12,7 @@ extern crate ffmpeg_next as ffmpeg;
extern crate ndarray;
use crate::chroma::ChromaDesc;
use crate::cue::CueInfo;
use crate::misc::LoudnessDesc;
#[cfg(doc)]
use crate::playlist;
use crate::playlist::{closest_to_first_song, dedup_playlist, euclidean_distance, DistanceMetric};
use crate::temporal::BPMDesc;
use crate::timbral::{SpectralDesc, ZeroCrossingRateDesc};
use crate::bliss_lib::{BlissError, BlissResult, SAMPLE_RATE};
@ -37,11 +33,9 @@ use ndarray::{arr1, Array1};
use std::convert::TryInto;
use std::fmt;
use std::path::Path;
use std::path::PathBuf;
use std::sync::mpsc;
use std::sync::mpsc::Receiver;
use std::thread as std_thread;
use std::time::Duration;
use strum::{EnumCount, IntoEnumIterator};
use strum_macros::{EnumCount, EnumIter};
@ -50,35 +44,12 @@ use strum_macros::{EnumCount, EnumIter};
/// Simple object used to represent a Song, with its path, analysis, and
/// other metadata (artist, genre...)
pub struct Song {
/// Song's provided file path
pub path: PathBuf,
/// Song's artist, read from the metadata
pub artist: Option<String>,
/// Song's title, read from the metadata
pub title: Option<String>,
/// Song's album name, read from the metadata
pub album: Option<String>,
/// Song's album's artist name, read from the metadata
pub album_artist: Option<String>,
/// Song's tracked number, read from the metadata
/// TODO normalize this into an integer
pub track_number: Option<String>,
/// Song's genre, read from the metadata (`""` if empty)
pub genre: Option<String>,
/// bliss analysis results
pub analysis: Analysis,
/// The song's duration
pub duration: Duration,
/// Version of the features the song was analyzed with.
/// A simple integer that is bumped every time a breaking change
/// is introduced in the features.
pub features_version: u16,
/// Populated only if the song was extracted from a larger audio file,
/// through the use of a CUE sheet.
/// By default, such a song's path would be
/// `path/to/cue_file.wav/CUE_TRACK00<track_number>`. Using this field,
/// you can change `song.path` to fit your needs.
pub cue_info: Option<CueInfo>,
}
#[derive(Debug, EnumIter, EnumCount)]
@ -218,96 +189,9 @@ impl Analysis {
Err(_) => None,
}
}
/// Compute distance between two analysis using a user-provided distance
/// metric. You most likely want to use `song.custom_distance` directly
/// rather than this function.
///
/// For this function to be integrated properly with the rest
/// of bliss' parts, it should be a valid distance metric, i.e.:
/// 1. For X, Y real vectors, d(X, Y) = 0 ⇔ X = Y
/// 2. For X, Y real vectors, d(X, Y) >= 0
/// 3. For X, Y real vectors, d(X, Y) = d(Y, X)
/// 4. For X, Y, Z real vectors d(X, Y) ≤ d(X + Z) + d(Z, Y)
///
/// Note that almost all distance metrics you will find obey these
/// properties, so don't sweat it too much.
pub fn custom_distance(&self, other: &Self, distance: impl DistanceMetric) -> f32 {
distance(&self.as_arr1(), &other.as_arr1())
}
}
impl Song {
#[allow(dead_code)]
/// Compute the distance between the current song and any given
/// Song.
///
/// The smaller the number, the closer the songs; usually more useful
/// if compared between several songs
/// (e.g. if song1.distance(song2) < song1.distance(song3), then song1 is
/// closer to song2 than it is to song3.
///
/// Currently uses the euclidean distance, but this can change in an
/// upcoming release if another metric performs better.
pub fn distance(&self, other: &Self) -> f32 {
self.analysis
.custom_distance(&other.analysis, euclidean_distance)
}
/// Compute distance between two songs using a user-provided distance
/// metric.
///
/// For this function to be integrated properly with the rest
/// of bliss' parts, it should be a valid distance metric, i.e.:
/// 1. For X, Y real vectors, d(X, Y) = 0 ⇔ X = Y
/// 2. For X, Y real vectors, d(X, Y) >= 0
/// 3. For X, Y real vectors, d(X, Y) = d(Y, X)
/// 4. For X, Y, Z real vectors d(X, Y) ≤ d(X + Z) + d(Z, Y)
///
/// Note that almost all distance metrics you will find obey these
/// properties, so don't sweat it too much.
pub fn custom_distance(&self, other: &Self, distance: impl DistanceMetric) -> f32 {
self.analysis.custom_distance(&other.analysis, distance)
}
/// Orders songs in `pool` by proximity to `self`, using the distance
/// metric `distance` to compute the order.
/// Basically return a playlist from songs in `pool`, starting
/// from `self`, using `distance` (some distance metrics can
/// be found in the [playlist] module).
///
/// Note that contrary to [Song::closest_from_pool], `self` is NOT added
/// to the beginning of the returned vector.
///
/// No deduplication is ran either; if you're looking for something easy
/// that works "out of the box", use [Song::closest_from_pool].
pub fn closest_from_pool_custom(
&self,
pool: Vec<Self>,
distance: impl DistanceMetric,
) -> Vec<Self> {
let mut pool = pool;
closest_to_first_song(self, &mut pool, distance);
pool
}
/// Order songs in `pool` by proximity to `self`.
/// Convenience method to return a playlist from songs in `pool`,
/// starting from `self`.
///
/// The distance is already chosen, deduplication is ran, and the first song
/// is added to the top of the playlist, to make everything easier.
///
/// If you want more control over which distance metric is chosen,
/// run deduplication manually, etc, use [Song::closest_from_pool_custom].
pub fn closest_from_pool(&self, pool: Vec<Self>) -> Vec<Self> {
let mut playlist = vec![self.to_owned()];
playlist.extend_from_slice(&pool);
closest_to_first_song(self, &mut playlist, euclidean_distance);
dedup_playlist(&mut playlist, None);
playlist
}
/// Returns a decoded [Song] given a file path, or an error if the song
/// could not be analyzed for some reason.
///
@ -325,20 +209,11 @@ impl Song {
/// decoding ([DecodingError](BlissError::DecodingError)) or an analysis
/// ([AnalysisError](BlissError::AnalysisError)) error.
pub fn from_path<P: AsRef<Path>>(path: P) -> BlissResult<Self> {
let raw_song = Song::decode(path.as_ref())?;
let samples = Song::decode(path.as_ref())?;
Ok(Song {
path: raw_song.path,
artist: raw_song.artist,
album_artist: raw_song.album_artist,
title: raw_song.title,
album: raw_song.album,
track_number: raw_song.track_number,
genre: raw_song.genre,
duration: raw_song.duration,
analysis: Song::analyze(&raw_song.sample_array)?,
analysis: Song::analyze(&samples)?,
features_version: FEATURES_VERSION,
cue_info: None,
})
}
@ -465,7 +340,7 @@ impl Song {
.unwrap()
}
pub(crate) fn decode(path: &Path) -> BlissResult<InternalSong> {
pub(crate) fn decode(path: &Path) -> BlissResult<Vec<f32>> {
ffmpeg::init().map_err(|e| {
BlissError::DecodingError(format!(
"ffmpeg init error while decoding file '{}': {:?}.",
@ -474,10 +349,7 @@ impl Song {
))
})?;
log::set_level(Level::Quiet);
let mut song = InternalSong {
path: path.into(),
..Default::default()
};
let mut ictx = ffmpeg::format::input(&path).map_err(|e| {
BlissError::DecodingError(format!(
"while opening format for file '{}': {:?}.",
@ -503,8 +375,7 @@ impl Song {
context.set_threading(Config {
kind: ThreadingType::Frame,
count: 0,
#[cfg(not(feature = "ffmpeg_6_0"))]
safe: true,
// safe: true,
});
let decoder = context.decoder().audio().map_err(|e| {
BlissError::DecodingError(format!(
@ -528,42 +399,7 @@ impl Song {
(decoder, input.index(), expected_sample_number)
};
let sample_array: Vec<f32> = Vec::with_capacity(expected_sample_number as usize);
if let Some(title) = ictx.metadata().get("title") {
song.title = match title {
"" => None,
t => Some(t.to_string()),
};
};
if let Some(artist) = ictx.metadata().get("artist") {
song.artist = match artist {
"" => None,
a => Some(a.to_string()),
};
};
if let Some(album) = ictx.metadata().get("album") {
song.album = match album {
"" => None,
a => Some(a.to_string()),
};
};
if let Some(genre) = ictx.metadata().get("genre") {
song.genre = match genre {
"" => None,
g => Some(g.to_string()),
};
};
if let Some(track_number) = ictx.metadata().get("track") {
song.track_number = match track_number {
"" => None,
t => Some(t.to_string()),
};
};
if let Some(album_artist) = ictx.metadata().get("album_artist") {
song.album_artist = match album_artist {
"" => None,
t => Some(t.to_string()),
};
};
let (empty_in_channel_layout, in_channel_layout) = {
if decoder.channel_layout() == ChannelLayout::empty() {
(true, ChannelLayout::default(decoder.channels().into()))
@ -604,8 +440,7 @@ impl Song {
path.display()
);
drop(tx);
song.sample_array = child.join().unwrap()?;
return Ok(song);
return Ok(child.join().unwrap()?);
}
Err(e) => warn!("error while decoding file '{}': {}", path.display(), e),
};
@ -643,8 +478,7 @@ impl Song {
path.display()
);
drop(tx);
song.sample_array = child.join().unwrap()?;
return Ok(song);
return Ok(child.join().unwrap()?);
}
Err(e) => warn!("error while decoding {}: {}", path.display(), e),
};
@ -666,26 +500,10 @@ impl Song {
}
drop(tx);
song.sample_array = child.join().unwrap()?;
let duration_seconds = song.sample_array.len() as f32 / SAMPLE_RATE as f32;
song.duration = Duration::from_nanos((duration_seconds * 1e9_f32).round() as u64);
Ok(song)
Ok(child.join().unwrap()?)
}
}
#[derive(Default, Debug)]
pub(crate) struct InternalSong {
pub path: PathBuf,
pub artist: Option<String>,
pub album_artist: Option<String>,
pub title: Option<String>,
pub album: Option<String>,
pub track_number: Option<String>,
pub genre: Option<String>,
pub duration: Duration,
pub sample_array: Vec<f32>,
}
fn resample_frame(
rx: Receiver<Audio>,
in_codec_format: Sample,
@ -831,42 +649,15 @@ mod tests {
}
fn _test_decode(path: &Path, expected_hash: &[u8]) {
let song = Song::decode(path).unwrap();
let samples = Song::decode(path).unwrap();
let mut hasher = Ripemd160::new();
for sample in song.sample_array.iter() {
for sample in samples.iter() {
hasher.update(sample.to_le_bytes().to_vec());
}
assert_eq!(expected_hash, hasher.finalize().as_slice());
}
#[test]
fn test_tags() {
let song = Song::decode(Path::new("data/s16_mono_22_5kHz.flac")).unwrap();
assert_eq!(song.artist, Some(String::from("David TMX")));
assert_eq!(
song.album_artist,
Some(String::from("David TMX - Album Artist"))
);
assert_eq!(song.title, Some(String::from("Renaissance")));
assert_eq!(song.album, Some(String::from("Renaissance")));
assert_eq!(song.track_number, Some(String::from("02")));
assert_eq!(song.genre, Some(String::from("Pop")));
// Test that there is less than 10ms of difference between what
// the song advertises and what we compute.
assert!((song.duration.as_millis() as f32 - 11070.).abs() < 10.);
}
#[test]
fn test_empty_tags() {
let song = Song::decode(Path::new("data/no_tags.flac")).unwrap();
assert_eq!(song.artist, None);
assert_eq!(song.title, None);
assert_eq!(song.album, None);
assert_eq!(song.track_number, None);
assert_eq!(song.genre, None);
}
#[test]
fn test_resample_multi() {
let path = Path::new("data/s32_stereo_44_1_kHz.flac");
@ -922,54 +713,23 @@ mod tests {
#[test]
fn test_decode_right_capacity_vec() {
let path = Path::new("data/s16_mono_22_5kHz.flac");
let song = Song::decode(&path).unwrap();
let sample_array = song.sample_array;
let samples = Song::decode(&path).unwrap();
assert_eq!(
sample_array.len() + SAMPLE_RATE as usize,
sample_array.capacity()
samples.len() + SAMPLE_RATE as usize,
samples.capacity()
);
let path = Path::new("data/s32_stereo_44_1_kHz.flac");
let song = Song::decode(&path).unwrap();
let sample_array = song.sample_array;
let samples = Song::decode(&path).unwrap();
assert_eq!(
sample_array.len() + SAMPLE_RATE as usize,
sample_array.capacity()
samples.len() + SAMPLE_RATE as usize,
samples.capacity()
);
let path = Path::new("data/capacity_fix.ogg");
let song = Song::decode(&path).unwrap();
let sample_array = song.sample_array;
assert!(sample_array.len() as f32 / sample_array.capacity() as f32 > 0.90);
assert!(sample_array.len() as f32 / (sample_array.capacity() as f32) < 1.);
}
#[test]
fn test_analysis_distance() {
let mut a = Song::default();
a.analysis = Analysis::new([
0.16391512, 0.11326739, 0.96868552, 0.8353934, 0.49867523, 0.76532606, 0.63448005,
0.82506196, 0.71457147, 0.62395476, 0.69680329, 0.9855766, 0.41369333, 0.13900452,
0.68001012, 0.11029723, 0.97192943, 0.57727861, 0.07994821, 0.88993185,
]);
let mut b = Song::default();
b.analysis = Analysis::new([
0.5075758, 0.36440256, 0.28888011, 0.43032829, 0.62387977, 0.61894916, 0.99676086,
0.11913155, 0.00640396, 0.15943407, 0.33829514, 0.34947174, 0.82927523, 0.18987604,
0.54437275, 0.22076826, 0.91232151, 0.29233168, 0.32846024, 0.04522147,
]);
assert_eq!(a.distance(&b), 1.9469079)
}
#[test]
fn test_analysis_distance_indiscernible() {
let mut a = Song::default();
a.analysis = Analysis::new([
1., 2., 3., 4., 5., 6., 7., 8., 9., 10., 11., 12., 13., 14., 15., 16., 17., 18., 19.,
20.,
]);
assert_eq!(a.distance(&a), 0.)
let samples = Song::decode(&path).unwrap();
assert!(samples.len() as f32 / samples.capacity() as f32 > 0.90);
assert!(samples.len() as f32 / (samples.capacity() as f32) < 1.);
}
#[test]
@ -1023,120 +783,4 @@ mod tests {
format!("{:?}", song.analysis),
);
}
fn dummy_distance(_: &Array1<f32>, _: &Array1<f32>) -> f32 {
0.
}
#[test]
fn test_custom_distance() {
let mut a = Song::default();
a.analysis = Analysis::new([
0.16391512, 0.11326739, 0.96868552, 0.8353934, 0.49867523, 0.76532606, 0.63448005,
0.82506196, 0.71457147, 0.62395476, 0.69680329, 0.9855766, 0.41369333, 0.13900452,
0.68001012, 0.11029723, 0.97192943, 0.57727861, 0.07994821, 0.88993185,
]);
let mut b = Song::default();
b.analysis = Analysis::new([
0.5075758, 0.36440256, 0.28888011, 0.43032829, 0.62387977, 0.61894916, 0.99676086,
0.11913155, 0.00640396, 0.15943407, 0.33829514, 0.34947174, 0.82927523, 0.18987604,
0.54437275, 0.22076826, 0.91232151, 0.29233168, 0.32846024, 0.04522147,
]);
assert_eq!(a.custom_distance(&b, dummy_distance), 0.);
}
#[test]
fn test_closest_from_pool() {
let song = Song {
path: Path::new("path-to-first").to_path_buf(),
analysis: Analysis::new([
1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1.,
]),
..Default::default()
};
let first_song_dupe = Song {
path: Path::new("path-to-dupe").to_path_buf(),
analysis: Analysis::new([
1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1., 1.,
]),
..Default::default()
};
let second_song = Song {
path: Path::new("path-to-second").to_path_buf(),
analysis: Analysis::new([
2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 1.9, 1., 1., 1.,
]),
..Default::default()
};
let third_song = Song {
path: Path::new("path-to-third").to_path_buf(),
analysis: Analysis::new([
2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2.5, 1., 1., 1.,
]),
..Default::default()
};
let fourth_song = Song {
path: Path::new("path-to-fourth").to_path_buf(),
analysis: Analysis::new([
2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 0., 1., 1., 1.,
]),
..Default::default()
};
let fifth_song = Song {
path: Path::new("path-to-fifth").to_path_buf(),
analysis: Analysis::new([
2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 2., 0., 1., 1., 1.,
]),
..Default::default()
};
let songs = vec![
song.to_owned(),
first_song_dupe.to_owned(),
second_song.to_owned(),
third_song.to_owned(),
fourth_song.to_owned(),
fifth_song.to_owned(),
];
let playlist = song.closest_from_pool(songs.to_owned());
assert_eq!(
playlist,
vec![
song.to_owned(),
second_song.to_owned(),
fourth_song.to_owned(),
third_song.to_owned(),
],
);
let playlist = song.closest_from_pool_custom(songs, euclidean_distance);
assert_eq!(
playlist,
vec![
song,
first_song_dupe,
second_song,
fourth_song,
fifth_song,
third_song
],
);
}
}
#[cfg(all(feature = "bench", test))]
mod bench {
extern crate test;
use crate::Song;
use std::path::Path;
use test::Bencher;
#[bench]
fn bench_resample_multi(b: &mut Bencher) {
let path = Path::new("./data/s32_stereo_44_1_kHz.flac");
b.iter(|| {
Song::decode(&path).unwrap();
});
}
}