Merge branch 'master' into github-mirror-master

Bump rust-ffmpeg to hopefully support rpis.

CHANGELOG.md

@@ -1,9 +1,13 @@
 #Changelog
 
+## bliss 0.6.11
+* Bump rust-ffmpeg to 6.1.1 to fix build for raspberry pis.
+
 ## bliss 0.6.10
 * Make the `analyze` function public, for people who don't want to use
   ffmpeg
-* Run `cargo update`
+* Run `cargo update`, bump ffmpeg to 6.1
+* Fix the library module erroring when wrong UTF-8 ends up in the database.
 
 ## bliss 0.6.9
 * Add a feature flag for compilation on raspberry pis.

Cargo.lock

@@ -2,6 +2,12 @@
 # It is not intended for manual editing.
 version = 3
 
+[[package]]
+name = "adler32"
+version = "1.2.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "aae1277d39aeec15cb388266ecc24b11c80469deae6067e17a1a7aa9e5c1f234"
+
 [[package]]
 name = "aho-corasick"
 version = "1.1.2"
@@ -89,8 +95,9 @@ dependencies = [
 
 [[package]]
 name = "bliss-rs"
-version = "0.6.9"
+version = "0.6.11"
 dependencies = [
+ "adler32",
  "anyhow",
  "bliss-audio-aubio-rs",
  "clap",
@@ -105,7 +112,6 @@ dependencies = [
  "neon",
  "noisy_float",
  "pretty_assertions",
- "ripemd",
  "rustfft",
  "serde",
  "serde_ini",
@@ -823,15 +829,6 @@ version = "1.0.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "194d8e591e405d1eecf28819740abed6d719d1a2db87fc0bcdedee9a26d55560"
 
-[[package]]
-name = "ripemd"
-version = "0.1.3"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "bd124222d17ad93a644ed9d011a40f4fb64aa54275c08cc216524a9ea82fb09f"
-dependencies = [
- "digest",
-]
-
 [[package]]
 name = "rustc-hash"
 version = "1.1.0"

Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "bliss-rs"
-version = "0.6.9"
+version = "0.6.11"
 build = "build.rs"
 authors = ["Polochon-street <polochonstreet@gmx.fr>"]
 edition = "2021"
@@ -10,7 +10,7 @@ homepage = "https://lelele.io/bliss.html"
 repository = "https://github.com/Polochon-street/bliss-rs"
 keywords = ["audio", "analysis", "MIR", "playlist", "similarity"]
 readme = "README.md"
-exclude = ["index.node"]
+exclude = ["data/", "index.node"]
 
 [lib]
 crate-type = ["rlib", "cdylib"]
@@ -24,12 +24,14 @@ no-default-features = true
 # Hopefully we'll be able to use the official aubio-rs at some point.
 bliss-audio-aubio-rs = { version = "0.2.1", features = ["static"] }
 crossbeam = "0.8.2"
-ffmpeg-next = { version = "6.0.0", features = ["static"] }
+ffmpeg-next = { version = "6.1.1", features = ["static"] }
 log = "0.4.17"
+# `rayon` is used only by `par_mapv_inplace` in chroma.rs.
+# TODO: is the speed gain that substantial?
 ndarray = { version = "0.15.6", features = ["rayon"] }
 ndarray-stats = "0.5.1"
 noisy_float = "0.2.0"
-ripemd = "0.1.3"
+adler32 = "1.0.2"
 rustfft = "6.1.0"
 thiserror = "1.0.40"
 strum = "0.24.1"

ci_check.sh

@@ -0,0 +1 @@
+cargo fmt -- --check && cargo clippy --examples --features=serde -- -D warnings && cargo build --verbose && cargo test --verbose && cargo test --verbose --examples && cargo +nightly-2023-02-16 bench --verbose --features=bench --no-run && cargo build --examples --verbose --features=serde

examples/distance.rs

@@ -13,8 +13,8 @@ fn main() -> Result<(), String> {
     let first_path = paths.next().ok_or("Help: ./distance <song1> <song2>")?;
     let second_path = paths.next().ok_or("Help: ./distance <song1> <song2>")?;
 
-    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 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();
@@ -25,8 +25,8 @@ fn main() -> Result<(), String> {
 
     println!(
         "d({:?}, {:?}) = {}",
-        song1.path,
-        song2.path,
+        &first_path,
+        &second_path,
         distance_squared.sqrt(),
     );
     Ok(())

src/chroma.rs

@@ -437,7 +437,7 @@ mod test {
     fn test_chroma_desc() {
         let song = Song::decode(Path::new("data/s16_mono_22_5kHz.flac")).unwrap();
         let mut chroma_desc = ChromaDesc::new(SAMPLE_RATE, 12);
-        chroma_desc.do_(&song.sample_array).unwrap();
+        chroma_desc.do_(&song).unwrap();
         let expected_values = vec![
             -0.35661936,
             -0.63578653,
@@ -457,9 +457,7 @@ mod test {
 
     #[test]
     fn test_chroma_stft_decode() {
-        let signal = Song::decode(Path::new("data/s16_mono_22_5kHz.flac"))
-            .unwrap()
-            .sample_array;
+        let signal = Song::decode(Path::new("data/s16_mono_22_5kHz.flac")).unwrap();
         let mut stft = stft(&signal, 8192, 2205);
 
         let file = File::open("data/chroma.npy").unwrap();
@@ -490,9 +488,7 @@ mod test {
 
     #[test]
     fn test_estimate_tuning_decode() {
-        let signal = Song::decode(Path::new("data/s16_mono_22_5kHz.flac"))
-            .unwrap()
-            .sample_array;
+        let signal = Song::decode(Path::new("data/s16_mono_22_5kHz.flac")).unwrap();
         let stft = stft(&signal, 8192, 2205);
 
         let tuning = estimate_tuning(22050, &stft, 8192, 0.01, 12).unwrap();

src/lib.rs

@@ -7,8 +7,8 @@ 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<()> {
@@ -59,13 +59,9 @@ 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) -> Result<([u8; 2], [u8; 80]), u8> {
-    return Ok(([0; 2], [0; 80]));
+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))
 }

src/misc.rs

@@ -70,7 +70,7 @@ mod tests {
     fn test_loudness() {
         let song = Song::decode(Path::new("data/s16_mono_22_5kHz.flac")).unwrap();
         let mut loudness_desc = LoudnessDesc::default();
-        for chunk in song.sample_array.chunks_exact(LoudnessDesc::WINDOW_SIZE) {
+        for chunk in song.chunks_exact(LoudnessDesc::WINDOW_SIZE) {
             loudness_desc.do_(&chunk);
         }
         let expected_values = vec![0.271263, 0.2577181];

src/song.rs

@@ -7,7 +7,6 @@
 //! For implementation of plug-ins for already existing audio players,
 //! a look at Library is instead recommended.
 
-extern crate crossbeam;
 extern crate ffmpeg_next as ffmpeg;
 extern crate ndarray;
 
@@ -19,7 +18,6 @@ use crate::bliss_lib::{BlissError, BlissResult, SAMPLE_RATE};
 use crate::bliss_lib::{CHANNELS, FEATURES_VERSION};
 use ::log::warn;
 use core::ops::Index;
-use crossbeam::thread;
 use ffmpeg_next::codec::threading::{Config, Type as ThreadingType};
 use ffmpeg_next::util::channel_layout::ChannelLayout;
 use ffmpeg_next::util::error::Error;
@@ -35,7 +33,7 @@ use std::fmt;
 use std::path::Path;
 use std::sync::mpsc;
 use std::sync::mpsc::Receiver;
-use std::thread as std_thread;
+use std::thread;
 use strum::{EnumCount, IntoEnumIterator};
 use strum_macros::{EnumCount, EnumIter};
 
@@ -234,16 +232,16 @@ impl Song {
      *
      * If you *do* want to use this with a song already decoded by yourself,
      * the sample format of `sample_array` should be f32le, one channel, and
-     * the sampling rate 22050 Hz. Anything other thant that will yield aberrant
+     * the sampling rate 22050 Hz. Anything other than that will yield aberrant
      * results.
      * To double-check that your sample array has the right format, you could run
-     * `ffmpeg -i path_to_your_song.flac -ar 22050 -ac 1 -c:a pcm_f32le -f hash -hash ripemd160 -`,
-     * which will give you the ripemd160 hash of the sample array if the song
-     * has been decoded properly. You can then compute the ripemd160 hash of your sample
-     * array (see `_test_decode` in the tests) and make sure both hashes are the same.
+     * `ffmpeg -i path_to_your_song.flac -ar 22050 -ac 1 -c:a pcm_f32le -f hash -hash addler32 -`,
+     * which will give you the addler32 checksum of the sample array if the song
+     * has been decoded properly. You can then compute the addler32 checksum of your sample
+     * array (see `_test_decode` in the tests) and make sure both are the same.
      *
      * (Running `ffmpeg -i path_to_your_song.flac -ar 22050 -ac 1 -c:a pcm_f32le` will simply give
-     * you the raw sample array as it should look like, if you're not into computing hashes)
+     * you the raw sample array as it should look like, if you're not into computing checksums)
      **/
     pub fn analyze(sample_array: &[f32]) -> BlissResult<Analysis> {
         let largest_window = vec![
@@ -261,8 +259,8 @@ impl Song {
             )));
         }
 
-        thread::scope(|s| {
-            let child_tempo: thread::ScopedJoinHandle<'_, BlissResult<f32>> = s.spawn(|_| {
+        thread::scope(|s| -> BlissResult<Analysis> {
+            let child_tempo = s.spawn(|| {
                 let mut tempo_desc = BPMDesc::new(SAMPLE_RATE)?;
                 let windows = sample_array
                     .windows(BPMDesc::WINDOW_SIZE)
@@ -274,17 +272,14 @@ impl Song {
                 Ok(tempo_desc.get_value())
             });
 
-            let child_chroma: thread::ScopedJoinHandle<'_, BlissResult<Vec<f32>>> = s.spawn(|_| {
+            let child_chroma = s.spawn(|| {
                 let mut chroma_desc = ChromaDesc::new(SAMPLE_RATE, 12);
                 chroma_desc.do_(sample_array)?;
                 Ok(chroma_desc.get_values())
             });
 
             #[allow(clippy::type_complexity)]
-            let child_timbral: thread::ScopedJoinHandle<
-                '_,
-                BlissResult<(Vec<f32>, Vec<f32>, Vec<f32>)>,
-            > = s.spawn(|_| {
+            let child_timbral = s.spawn(|| {
                 let mut spectral_desc = SpectralDesc::new(SAMPLE_RATE)?;
                 let windows = sample_array
                     .windows(SpectralDesc::WINDOW_SIZE)
@@ -298,14 +293,13 @@ impl Song {
                 Ok((centroid, rolloff, flatness))
             });
 
-            let child_zcr: thread::ScopedJoinHandle<'_, BlissResult<f32>> = s.spawn(|_| {
+            let child_zcr = s.spawn(|| {
                 let mut zcr_desc = ZeroCrossingRateDesc::default();
                 zcr_desc.do_(sample_array);
                 Ok(zcr_desc.get_value())
             });
 
-            let child_loudness: thread::ScopedJoinHandle<'_, BlissResult<Vec<f32>>> =
-                s.spawn(|_| {
+            let child_loudness = s.spawn(|| {
                 let mut loudness_desc = LoudnessDesc::default();
                 let windows = sample_array.chunks(LoudnessDesc::WINDOW_SIZE);
 
@@ -337,7 +331,6 @@ impl Song {
             })?;
             Ok(Analysis::new(array))
         })
-        .unwrap()
     }
 
     pub(crate) fn decode(path: &Path) -> BlissResult<Vec<f32>> {
@@ -412,7 +405,7 @@ impl Song {
         let (tx, rx) = mpsc::channel();
         let in_codec_format = decoder.format();
         let in_codec_rate = decoder.rate();
-        let child = std_thread::spawn(move || {
+        let child = thread::spawn(move || {
             resample_frame(
                 rx,
                 in_codec_format,
@@ -598,8 +591,8 @@ fn push_to_sample_array(frame: &ffmpeg::frame::Audio, sample_array: &mut Vec<f32
 #[cfg(test)]
 mod tests {
     use super::*;
+    use adler32::RollingAdler32;
     use pretty_assertions::assert_eq;
-    use ripemd::{Digest, Ripemd160};
     use std::path::Path;
 
     #[test]
@@ -648,34 +641,28 @@ mod tests {
         assert_eq!(FEATURES_VERSION, song.features_version);
     }
 
-    fn _test_decode(path: &Path, expected_hash: &[u8]) {
+    fn _test_decode(path: &Path, expected_hash: u32) {
         let samples = Song::decode(path).unwrap();
-        let mut hasher = Ripemd160::new();
+        let mut hasher = RollingAdler32::new();
         for sample in samples.iter() {
-            hasher.update(sample.to_le_bytes().to_vec());
+            hasher.update_buffer(&sample.to_le_bytes());
         }
 
-        assert_eq!(expected_hash, hasher.finalize().as_slice());
+        assert_eq!(expected_hash, hasher.hash());
     }
 
     #[test]
     fn test_resample_multi() {
         let path = Path::new("data/s32_stereo_44_1_kHz.flac");
-        let expected_hash = [
-            0xc5, 0xf8, 0x23, 0xce, 0x63, 0x2c, 0xf4, 0xa0, 0x72, 0x66, 0xbb, 0x49, 0xad, 0x84,
-            0xb6, 0xea, 0x48, 0x48, 0x9c, 0x50,
-        ];
-        _test_decode(&path, &expected_hash);
+        let expected_hash = 0xbbcba1cf;
+        _test_decode(&path, expected_hash);
     }
 
     #[test]
     fn test_resample_stereo() {
         let path = Path::new("data/s16_stereo_22_5kHz.flac");
-        let expected_hash = [
-            0x24, 0xed, 0x45, 0x58, 0x06, 0xbf, 0xfb, 0x05, 0x57, 0x5f, 0xdc, 0x4d, 0xb4, 0x9b,
-            0xa5, 0x2b, 0x05, 0x56, 0x10, 0x4f,
-        ];
-        _test_decode(&path, &expected_hash);
+        let expected_hash = 0x1d7b2d6d;
+        _test_decode(&path, expected_hash);
     }
 
     #[test]
@@ -683,12 +670,9 @@ mod tests {
         let path = Path::new("data/s16_mono_22_5kHz.flac");
         // Obtained through
         // ffmpeg -i data/s16_mono_22_5kHz.flac -ar 22050 -ac 1 -c:a pcm_f32le
-        // -f hash -hash ripemd160 -
-        let expected_hash = [
-            0x9d, 0x95, 0xa5, 0xf2, 0xd2, 0x9c, 0x68, 0xe8, 0x8a, 0x70, 0xcd, 0xf3, 0x54, 0x2c,
-            0x5b, 0x45, 0x98, 0xb4, 0xf3, 0xb4,
-        ];
-        _test_decode(&path, &expected_hash);
+        // -f hash -hash addler32 -
+        let expected_hash = 0x5e01930b;
+        _test_decode(&path, expected_hash);
     }
 
     #[test]
@@ -696,12 +680,9 @@ mod tests {
         let path = Path::new("data/s32_stereo_44_1_kHz.mp3");
         // Obtained through
         // ffmpeg -i data/s16_mono_22_5kHz.mp3 -ar 22050 -ac 1 -c:a pcm_f32le
-        // -f hash -hash ripemd160 -
-        let expected_hash = [
-            0x28, 0x25, 0x6b, 0x7b, 0x6e, 0x37, 0x1c, 0xcf, 0xc7, 0x06, 0xdf, 0x62, 0x8c, 0x0e,
-            0x91, 0xf7, 0xd6, 0x1f, 0xac, 0x5b,
-        ];
-        _test_decode(&path, &expected_hash);
+        // -f hash -hash addler32 -
+        let expected_hash = 0x69ca6906;
+        _test_decode(&path, expected_hash);
     }
 
     #[test]
@@ -757,11 +738,8 @@ mod tests {
 
     #[test]
     fn test_decode_wav() {
-        let expected_hash = [
-            0xf0, 0xe0, 0x85, 0x4e, 0xf6, 0x53, 0x76, 0xfa, 0x7a, 0xa5, 0x65, 0x76, 0xf9, 0xe1,
-            0xe8, 0xe0, 0x81, 0xc8, 0xdc, 0x61,
-        ];
-        _test_decode(Path::new("data/piano.wav"), &expected_hash);
+        let expected_hash = 0xde831e82;
+        _test_decode(Path::new("data/piano.wav"), expected_hash);
     }
 
     #[test]

src/temporal.rs

@@ -101,7 +101,7 @@ mod tests {
     fn test_tempo_real() {
         let song = Song::decode(Path::new("data/s16_mono_22_5kHz.flac")).unwrap();
         let mut tempo_desc = BPMDesc::new(SAMPLE_RATE).unwrap();
-        for chunk in song.sample_array.chunks_exact(BPMDesc::HOP_SIZE) {
+        for chunk in song.chunks_exact(BPMDesc::HOP_SIZE) {
             tempo_desc.do_(&chunk).unwrap();
         }
         assert!(0.01 > (0.378605 - tempo_desc.get_value()).abs());

src/timbral.rs

@@ -283,7 +283,7 @@ mod tests {
     fn test_zcr() {
         let song = Song::decode(Path::new("data/s16_mono_22_5kHz.flac")).unwrap();
         let mut zcr_desc = ZeroCrossingRateDesc::default();
-        for chunk in song.sample_array.chunks_exact(SpectralDesc::HOP_SIZE) {
+        for chunk in song.chunks_exact(SpectralDesc::HOP_SIZE) {
             zcr_desc.do_(&chunk);
         }
         assert!(0.001 > (-0.85036 - zcr_desc.get_value()).abs());
@@ -303,13 +303,14 @@ mod tests {
             assert!(0.0000001 > (expected - actual).abs());
         }
 
-        let song = Song::decode(Path::new("data/white_noise.flac")).unwrap();
+        let song = Song::decode(Path::new("data/white_noise.mp3")).unwrap();
         let mut spectral_desc = SpectralDesc::new(22050).unwrap();
-        for chunk in song.sample_array.chunks_exact(SpectralDesc::HOP_SIZE) {
+        for chunk in song.chunks_exact(SpectralDesc::HOP_SIZE) {
             spectral_desc.do_(&chunk).unwrap();
         }
+        println!("{:?}", spectral_desc.get_flatness());
         // White noise - as close to 1 as possible
-        let expected_values = vec![0.6706717, -0.9685736];
+        let expected_values = vec![0.5785303, -0.9426308];
         for (expected, actual) in expected_values
             .iter()
             .zip(spectral_desc.get_flatness().iter())
@@ -322,7 +323,7 @@ mod tests {
     fn test_spectral_flatness() {
         let song = Song::decode(Path::new("data/s16_mono_22_5kHz.flac")).unwrap();
         let mut spectral_desc = SpectralDesc::new(SAMPLE_RATE).unwrap();
-        for chunk in song.sample_array.chunks_exact(SpectralDesc::HOP_SIZE) {
+        for chunk in song.chunks_exact(SpectralDesc::HOP_SIZE) {
             spectral_desc.do_(&chunk).unwrap();
         }
         // Spectral flatness mean value computed here with phase vocoder before normalization: 0.111949615
@@ -352,7 +353,7 @@ mod tests {
 
         let song = Song::decode(Path::new("data/tone_11080Hz.flac")).unwrap();
         let mut spectral_desc = SpectralDesc::new(SAMPLE_RATE).unwrap();
-        for chunk in song.sample_array.chunks_exact(SpectralDesc::HOP_SIZE) {
+        for chunk in song.chunks_exact(SpectralDesc::HOP_SIZE) {
             spectral_desc.do_(&chunk).unwrap();
         }
         let expected_values = vec![0.9967681, -0.99615175];
@@ -368,7 +369,7 @@ mod tests {
     fn test_spectral_roll_off() {
         let song = Song::decode(Path::new("data/s16_mono_22_5kHz.flac")).unwrap();
         let mut spectral_desc = SpectralDesc::new(SAMPLE_RATE).unwrap();
-        for chunk in song.sample_array.chunks_exact(SpectralDesc::HOP_SIZE) {
+        for chunk in song.chunks_exact(SpectralDesc::HOP_SIZE) {
             spectral_desc.do_(&chunk).unwrap();
         }
         let expected_values = vec![-0.6326486, -0.7260933];
@@ -386,7 +387,7 @@ mod tests {
     fn test_spectral_centroid() {
         let song = Song::decode(Path::new("data/s16_mono_22_5kHz.flac")).unwrap();
         let mut spectral_desc = SpectralDesc::new(SAMPLE_RATE).unwrap();
-        for chunk in song.sample_array.chunks_exact(SpectralDesc::HOP_SIZE) {
+        for chunk in song.chunks_exact(SpectralDesc::HOP_SIZE) {
             spectral_desc.do_(&chunk).unwrap();
         }
         // Spectral centroid mean value computed here with phase vocoder before normalization: 1354.2273
@@ -415,7 +416,7 @@ mod tests {
         }
         let song = Song::decode(Path::new("data/tone_11080Hz.flac")).unwrap();
         let mut spectral_desc = SpectralDesc::new(SAMPLE_RATE).unwrap();
-        for chunk in song.sample_array.chunks_exact(SpectralDesc::HOP_SIZE) {
+        for chunk in song.chunks_exact(SpectralDesc::HOP_SIZE) {
             spectral_desc.do_(&chunk).unwrap();
         }
         let expected_values = vec![0.97266, -0.9609926];

src/utils.rs

@@ -498,7 +498,7 @@ mod tests {
 
         let song = Song::decode(Path::new("data/piano.flac")).unwrap();
 
-        let stft = stft(&song.sample_array, 2048, 512);
+        let stft = stft(&song, 2048, 512);
 
         assert!(!stft.is_empty() && !expected_stft.is_empty());
         for (expected, actual) in expected_stft.iter().zip(stft.iter()) {