Merge pull request #660 from roderickvd/hi-res-volume-control

High-resolution volume control and normalisation

Author: sashahilton00

Cargo.lock

@@ -22,9 +22,10 @@ log = "0.4"
 num-bigint = "0.3"
 num-traits = "0.2"
 tempfile = "3.1"
+zerocopy = "0.3"
 
-librespot-tremor = { version = "0.2.0", optional = true }
-vorbis = { version ="0.0.14", optional = true }
+librespot-tremor = { version = "0.2", optional = true }
+vorbis = { version ="0.0", optional = true }
 
 [features]
 with-tremor = ["librespot-tremor"]

audio/Cargo.toml

@@ -0,0 +1,59 @@
+use zerocopy::AsBytes;
+
+#[derive(AsBytes, Copy, Clone, Debug)]
+#[allow(non_camel_case_types)]
+#[repr(transparent)]
+pub struct i24([u8; 3]);
+impl i24 {
+    fn pcm_from_i32(sample: i32) -> Self {
+        // drop the least significant byte
+        let [a, b, c, _d] = (sample >> 8).to_le_bytes();
+        i24([a, b, c])
+    }
+}
+
+// Losslessly represent [-1.0, 1.0] to [$type::MIN, $type::MAX] while maintaining DC linearity.
+macro_rules! convert_samples_to {
+    ($type: ident, $samples: expr) => {
+        convert_samples_to!($type, $samples, 0)
+    };
+    ($type: ident, $samples: expr, $drop_bits: expr) => {
+        $samples
+            .iter()
+            .map(|sample| {
+                // Losslessly represent [-1.0, 1.0] to [$type::MIN, $type::MAX]
+                // while maintaining DC linearity. There is nothing to be gained
+                // by doing this in f64, as the significand of a f32 is 24 bits,
+                // just like the maximum bit depth we are converting to.
+                let int_value = *sample * (std::$type::MAX as f32 + 0.5) - 0.5;
+
+                // Casting floats to ints truncates by default, which results
+                // in larger quantization error than rounding arithmetically.
+                // Flooring is faster, but again with larger error.
+                int_value.round() as $type >> $drop_bits
+            })
+            .collect()
+    };
+}
+
+pub struct SamplesConverter {}
+impl SamplesConverter {
+    pub fn to_s32(samples: &[f32]) -> Vec<i32> {
+        convert_samples_to!(i32, samples)
+    }
+
+    pub fn to_s24(samples: &[f32]) -> Vec<i32> {
+        convert_samples_to!(i32, samples, 8)
+    }
+
+    pub fn to_s24_3(samples: &[f32]) -> Vec<i24> {
+        Self::to_s32(samples)
+            .iter()
+            .map(|sample| i24::pcm_from_i32(*sample))
+            .collect()
+    }
+
+    pub fn to_s16(samples: &[f32]) -> Vec<i16> {
+        convert_samples_to!(i16, samples)
+    }
+}

audio/src/convert.rs

@@ -37,8 +37,11 @@ where
         use self::lewton::VorbisError::BadAudio;
         use self::lewton::VorbisError::OggError;
         loop {
-            match self.0.read_dec_packet_itl() {
-                Ok(Some(packet)) => return Ok(Some(AudioPacket::Samples(packet))),
+            match self
+                .0
+                .read_dec_packet_generic::<lewton::samples::InterleavedSamples<f32>>()
+            {
+                Ok(Some(packet)) => return Ok(Some(AudioPacket::Samples(packet.samples))),
                 Ok(None) => return Ok(None),
 
                 Err(BadAudio(AudioIsHeader)) => (),

audio/src/lewton_decoder.rs

@@ -13,6 +13,7 @@ extern crate tempfile;
 
 extern crate librespot_core;
 
+mod convert;
 mod decrypt;
 mod fetch;
 
@@ -24,6 +25,7 @@ mod passthrough_decoder;
 
 mod range_set;
 
+pub use convert::{i24, SamplesConverter};
 pub use decrypt::AudioDecrypt;
 pub use fetch::{AudioFile, AudioFileOpen, StreamLoaderController};
 pub use fetch::{
@@ -33,12 +35,12 @@ pub use fetch::{
 use std::fmt;
 
 pub enum AudioPacket {
-    Samples(Vec<i16>),
+    Samples(Vec<f32>),
     OggData(Vec<u8>),
 }
 
 impl AudioPacket {
-    pub fn samples(&self) -> &[i16] {
+    pub fn samples(&self) -> &[f32] {
         match self {
             AudioPacket::Samples(s) => s,
             AudioPacket::OggData(_) => panic!("can't return OggData on samples"),

audio/src/lib.rs

@@ -39,7 +39,18 @@ where
     fn next_packet(&mut self) -> Result<Option<AudioPacket>, AudioError> {
         loop {
             match self.0.packets().next() {
-                Some(Ok(packet)) => return Ok(Some(AudioPacket::Samples(packet.data))),
+                Some(Ok(packet)) => {
+                    // Losslessly represent [-32768, 32767] to [-1.0, 1.0] while maintaining DC linearity.
+                    return Ok(Some(AudioPacket::Samples(
+                        packet
+                            .data
+                            .iter()
+                            .map(|sample| {
+                                ((*sample as f64 + 0.5) / (std::i16::MAX as f64 + 0.5)) as f32
+                            })
+                            .collect(),
+                    )));
+                }
                 None => return Ok(None),
 
                 Some(Err(vorbis::VorbisError::Hole)) => (),

audio/src/libvorbis_decoder.rs

@@ -5,7 +5,7 @@ use librespot::core::authentication::Credentials;
 use librespot::core::config::SessionConfig;
 use librespot::core::session::Session;
 use librespot::core::spotify_id::SpotifyId;
-use librespot::playback::config::PlayerConfig;
+use librespot::playback::config::{AudioFormat, PlayerConfig};
 
 use librespot::playback::audio_backend;
 use librespot::playback::player::Player;
@@ -16,6 +16,7 @@ fn main() {
 
     let session_config = SessionConfig::default();
     let player_config = PlayerConfig::default();
+    let audio_format = AudioFormat::default();
 
     let args: Vec<_> = env::args().collect();
     if args.len() != 4 {
@@ -35,7 +36,7 @@ fn main() {
         .unwrap();
 
     let (mut player, _) = Player::new(player_config, session.clone(), None, move || {
-        (backend)(None)
+        (backend)(None, audio_format)
     });
 
     player.load(track, true, 0);

examples/play.rs

@@ -23,19 +23,19 @@ log = "0.4"
 byteorder = "1.3"
 shell-words = "1.0.0"
 
-alsa            = { version = "0.4", optional = true }
+alsa            = { version = "0.5", optional = true }
 portaudio-rs    = { version = "0.3", optional = true }
-libpulse-binding        = { version = "2.13", optional = true, default-features = false }
-libpulse-simple-binding = { version = "2.13", optional = true, default-features = false }
+libpulse-binding        = { version = "2", optional = true, default-features = false }
+libpulse-simple-binding = { version = "2", optional = true, default-features = false }
 jack            = { version = "0.6", optional = true }
 libc            = { version = "0.2", optional = true }
 rodio           = { version = "0.13", optional = true, default-features = false }
 cpal            = { version = "0.13", optional = true }
-sdl2            = { version = "0.34", optional = true }
+sdl2            = { version = "0.34.3", optional = true }
 gstreamer       = { version = "0.16", optional = true }
 gstreamer-app   = { version = "0.16", optional = true }
 glib            = { version = "0.10", optional = true }
-zerocopy        = { version = "0.3", optional = true }
+zerocopy        = { version = "0.3" }
 
 [features]
 alsa-backend = ["alsa"]
@@ -45,4 +45,4 @@ jackaudio-backend = ["jack"]
 rodiojack-backend = ["rodio", "cpal/jack"]
 rodio-backend = ["rodio", "cpal"]
 sdl-backend = ["sdl2"]
-gstreamer-backend = ["gstreamer", "gstreamer-app", "glib", "zerocopy"]
+gstreamer-backend = ["gstreamer", "gstreamer-app", "glib"]

playback/Cargo.toml

@@ -1,5 +1,7 @@
-use super::{Open, Sink};
+use super::{Open, Sink, SinkAsBytes};
 use crate::audio::AudioPacket;
+use crate::config::AudioFormat;
+use crate::player::{NUM_CHANNELS, SAMPLES_PER_SECOND, SAMPLE_RATE};
 use alsa::device_name::HintIter;
 use alsa::pcm::{Access, Format, Frames, HwParams, PCM};
 use alsa::{Direction, Error, ValueOr};
@@ -8,13 +10,14 @@ use std::ffi::CString;
 use std::io;
 use std::process::exit;
 
-const PREFERED_PERIOD_SIZE: Frames = 5512; // Period of roughly 125ms
+const BUFFERED_LATENCY: f32 = 0.125; // seconds
 const BUFFERED_PERIODS: Frames = 4;
 
 pub struct AlsaSink {
     pcm: Option<PCM>,
+    format: AudioFormat,
     device: String,
-    buffer: Vec<i16>,
+    buffer: Vec<u8>,
 }
 
 fn list_outputs() {
@@ -34,23 +37,27 @@ fn list_outputs() {
     }
 }
 
-fn open_device(dev_name: &str) -> Result<(PCM, Frames), Box<Error>> {
+fn open_device(dev_name: &str, format: AudioFormat) -> Result<(PCM, Frames), Box<Error>> {
     let pcm = PCM::new(dev_name, Direction::Playback, false)?;
-    let mut period_size = PREFERED_PERIOD_SIZE;
+    let alsa_format = match format {
+        AudioFormat::F32 => Format::float(),
+        AudioFormat::S32 => Format::s32(),
+        AudioFormat::S24 => Format::s24(),
+        AudioFormat::S24_3 => Format::S243LE,
+        AudioFormat::S16 => Format::s16(),
+    };
+
     // http://www.linuxjournal.com/article/6735?page=0,1#N0x19ab2890.0x19ba78d8
     // latency = period_size * periods / (rate * bytes_per_frame)
-    // For 16 Bit stereo data, one frame has a length of four bytes.
-    // 500ms  = buffer_size / (44100 * 4)
-    // buffer_size_bytes = 0.5 * 44100 / 4
-    // buffer_size_frames = 0.5 * 44100 = 22050
+    // For stereo samples encoded as 32-bit float, one frame has a length of eight bytes.
+    let mut period_size = ((SAMPLES_PER_SECOND * format.size() as u32) as f32
+        * (BUFFERED_LATENCY / BUFFERED_PERIODS as f32)) as Frames;
     {
-        // Set hardware parameters: 44100 Hz / Stereo / 16 bit
         let hwp = HwParams::any(&pcm)?;
-
         hwp.set_access(Access::RWInterleaved)?;
-        hwp.set_format(Format::s16())?;
-        hwp.set_rate(44100, ValueOr::Nearest)?;
-        hwp.set_channels(2)?;
+        hwp.set_format(alsa_format)?;
+        hwp.set_rate(SAMPLE_RATE, ValueOr::Nearest)?;
+        hwp.set_channels(NUM_CHANNELS as u32)?;
         period_size = hwp.set_period_size_near(period_size, ValueOr::Greater)?;
         hwp.set_buffer_size_near(period_size * BUFFERED_PERIODS)?;
         pcm.hw_params(&hwp)?;
@@ -64,12 +71,12 @@ fn open_device(dev_name: &str) -> Result<(PCM, Frames), Box<Error>> {
 }
 
 impl Open for AlsaSink {
-    fn open(device: Option<String>) -> AlsaSink {
-        info!("Using alsa sink");
+    fn open(device: Option<String>, format: AudioFormat) -> Self {
+        info!("Using Alsa sink with format: {:?}", format);
 
         let name = match device.as_ref().map(AsRef::as_ref) {
             Some("?") => {
-                println!("Listing available alsa outputs");
+                println!("Listing available Alsa outputs:");
                 list_outputs();
                 exit(0)
             }
@@ -78,8 +85,9 @@ impl Open for AlsaSink {
         }
         .to_string();
 
-        AlsaSink {
+        Self {
             pcm: None,
+            format: format,
             device: name,
             buffer: vec![],
         }
@@ -89,12 +97,14 @@ impl Open for AlsaSink {
 impl Sink for AlsaSink {
     fn start(&mut self) -> io::Result<()> {
         if self.pcm.is_none() {
-            let pcm = open_device(&self.device);
+            let pcm = open_device(&self.device, self.format);
             match pcm {
                 Ok((p, period_size)) => {
                     self.pcm = Some(p);
                     // Create a buffer for all samples for a full period
-                    self.buffer = Vec::with_capacity((period_size * 2) as usize);
+                    self.buffer = Vec::with_capacity(
+                        period_size as usize * BUFFERED_PERIODS as usize * self.format.size(),
+                    );
                 }
                 Err(e) => {
                     error!("Alsa error PCM open {}", e);
@@ -111,23 +121,22 @@ impl Sink for AlsaSink {
 
     fn stop(&mut self) -> io::Result<()> {
         {
-            let pcm = self.pcm.as_mut().unwrap();
             // Write any leftover data in the period buffer
             // before draining the actual buffer
-            let io = pcm.io_i16().unwrap();
-            match io.writei(&self.buffer[..]) {
-                Ok(_) => (),
-                Err(err) => pcm.try_recover(err, false).unwrap(),
-            }
+            self.write_bytes(&[]).expect("could not flush buffer");
+            let pcm = self.pcm.as_mut().unwrap();
             pcm.drain().unwrap();
         }
         self.pcm = None;
         Ok(())
     }
 
-    fn write(&mut self, packet: &AudioPacket) -> io::Result<()> {
+    sink_as_bytes!();
+}
+
+impl SinkAsBytes for AlsaSink {
+    fn write_bytes(&mut self, data: &[u8]) -> io::Result<()> {
         let mut processed_data = 0;
-        let data = packet.samples();
         while processed_data < data.len() {
             let data_to_buffer = min(
                 self.buffer.capacity() - self.buffer.len(),
@@ -137,16 +146,24 @@ impl Sink for AlsaSink {
                 .extend_from_slice(&data[processed_data..processed_data + data_to_buffer]);
             processed_data += data_to_buffer;
             if self.buffer.len() == self.buffer.capacity() {
-                let pcm = self.pcm.as_mut().unwrap();
-                let io = pcm.io_i16().unwrap();
-                match io.writei(&self.buffer) {
-                    Ok(_) => (),
-                    Err(err) => pcm.try_recover(err, false).unwrap(),
-                }
+                self.write_buf().expect("could not append to buffer");
                 self.buffer.clear();
             }
         }
 
         Ok(())
     }
 }
+
+impl AlsaSink {
+    fn write_buf(&mut self) -> io::Result<()> {
+        let pcm = self.pcm.as_mut().unwrap();
+        let io = pcm.io_bytes();
+        match io.writei(&self.buffer) {
+            Ok(_) => (),
+            Err(err) => pcm.try_recover(err, false).unwrap(),
+        };
+
+        Ok(())
+    }
+}