1use anyhow::{Context as _, Result};
2use collections::HashMap;
3use gpui::{App, BackgroundExecutor, BorrowAppContext, Global};
4
5#[cfg(not(any(all(target_os = "windows", target_env = "gnu"), target_os = "freebsd")))]
6mod non_windows_and_freebsd_deps {
7 pub(super) use gpui::AsyncApp;
8 pub(super) use libwebrtc::native::apm;
9 pub(super) use log::info;
10 pub(super) use parking_lot::Mutex;
11 pub(super) use rodio::cpal::Sample;
12 pub(super) use rodio::source::LimitSettings;
13 pub(super) use std::sync::Arc;
14}
15
16#[cfg(not(any(all(target_os = "windows", target_env = "gnu"), target_os = "freebsd")))]
17use non_windows_and_freebsd_deps::*;
18
19use rodio::{
20 Decoder, OutputStream, OutputStreamBuilder, Source, mixer::Mixer, nz, source::Buffered,
21};
22use settings::Settings;
23use std::{io::Cursor, num::NonZero, path::PathBuf, sync::atomic::Ordering, time::Duration};
24use util::ResultExt;
25
26mod audio_settings;
27mod replays;
28mod rodio_ext;
29pub use audio_settings::AudioSettings;
30pub use rodio_ext::RodioExt;
31
32use crate::audio_settings::LIVE_SETTINGS;
33
34// We are migrating to 16kHz sample rate from 48kHz. In the future
35// once we are reasonably sure most users have upgraded we will
36// remove the LEGACY parameters.
37//
38// We migrate to 16kHz because it is sufficient for speech and required
39// by the denoiser and future Speech to Text layers.
40pub const SAMPLE_RATE: NonZero<u32> = nz!(16000);
41pub const CHANNEL_COUNT: NonZero<u16> = nz!(1);
42pub const BUFFER_SIZE: usize = // echo canceller and livekit want 10ms of audio
43 (SAMPLE_RATE.get() as usize / 100) * CHANNEL_COUNT.get() as usize;
44
45pub const LEGACY_SAMPLE_RATE: NonZero<u32> = nz!(48000);
46pub const LEGACY_CHANNEL_COUNT: NonZero<u16> = nz!(2);
47
48pub const REPLAY_DURATION: Duration = Duration::from_secs(30);
49
50pub fn init(cx: &mut App) {
51 AudioSettings::register(cx);
52 LIVE_SETTINGS.initialize(cx);
53}
54
55#[derive(Debug, Copy, Clone, Eq, Hash, PartialEq)]
56pub enum Sound {
57 Joined,
58 Leave,
59 Mute,
60 Unmute,
61 StartScreenshare,
62 StopScreenshare,
63 AgentDone,
64}
65
66impl Sound {
67 fn file(&self) -> &'static str {
68 match self {
69 Self::Joined => "joined_call",
70 Self::Leave => "leave_call",
71 Self::Mute => "mute",
72 Self::Unmute => "unmute",
73 Self::StartScreenshare => "start_screenshare",
74 Self::StopScreenshare => "stop_screenshare",
75 Self::AgentDone => "agent_done",
76 }
77 }
78}
79
80pub struct Audio {
81 output_handle: Option<OutputStream>,
82 output_mixer: Option<Mixer>,
83 #[cfg(not(any(all(target_os = "windows", target_env = "gnu"), target_os = "freebsd")))]
84 pub echo_canceller: Arc<Mutex<apm::AudioProcessingModule>>,
85 source_cache: HashMap<Sound, Buffered<Decoder<Cursor<Vec<u8>>>>>,
86 replays: replays::Replays,
87}
88
89impl Default for Audio {
90 fn default() -> Self {
91 Self {
92 output_handle: Default::default(),
93 output_mixer: Default::default(),
94 #[cfg(not(any(
95 all(target_os = "windows", target_env = "gnu"),
96 target_os = "freebsd"
97 )))]
98 echo_canceller: Arc::new(Mutex::new(apm::AudioProcessingModule::new(
99 true, false, false, false,
100 ))),
101 source_cache: Default::default(),
102 replays: Default::default(),
103 }
104 }
105}
106
107impl Global for Audio {}
108
109impl Audio {
110 fn ensure_output_exists(&mut self) -> Result<&Mixer> {
111 #[cfg(debug_assertions)]
112 log::warn!(
113 "Audio does not sound correct without optimizations. Use a release build to debug audio issues"
114 );
115
116 if self.output_handle.is_none() {
117 let output_handle = OutputStreamBuilder::open_default_stream()
118 .context("Could not open default output stream")?;
119 info!("Output stream: {:?}", output_handle);
120 self.output_handle = Some(output_handle);
121 if let Some(output_handle) = &self.output_handle {
122 let (mixer, source) = rodio::mixer::mixer(CHANNEL_COUNT, SAMPLE_RATE);
123 // or the mixer will end immediately as its empty.
124 mixer.add(rodio::source::Zero::new(CHANNEL_COUNT, SAMPLE_RATE));
125 self.output_mixer = Some(mixer);
126
127 // The webrtc apm is not yet compiling for windows & freebsd
128 #[cfg(not(any(
129 any(all(target_os = "windows", target_env = "gnu")),
130 target_os = "freebsd"
131 )))]
132 let echo_canceller = Arc::clone(&self.echo_canceller);
133 #[cfg(not(any(
134 any(all(target_os = "windows", target_env = "gnu")),
135 target_os = "freebsd"
136 )))]
137 let source = source.inspect_buffer::<BUFFER_SIZE, _>(move |buffer| {
138 let mut buf: [i16; _] = buffer.map(|s| s.to_sample());
139 echo_canceller
140 .lock()
141 .process_reverse_stream(
142 &mut buf,
143 SAMPLE_RATE.get() as i32,
144 CHANNEL_COUNT.get().into(),
145 )
146 .expect("Audio input and output threads should not panic");
147 });
148 output_handle.mixer().add(source);
149 }
150 }
151
152 Ok(self
153 .output_mixer
154 .as_ref()
155 .expect("we only get here if opening the outputstream succeeded"))
156 }
157
158 pub fn save_replays(
159 &self,
160 executor: BackgroundExecutor,
161 ) -> gpui::Task<anyhow::Result<(PathBuf, Duration)>> {
162 self.replays.replays_to_tar(executor)
163 }
164
165 #[cfg(not(any(all(target_os = "windows", target_env = "gnu"), target_os = "freebsd")))]
166 pub fn open_microphone(voip_parts: VoipParts) -> anyhow::Result<impl Source> {
167 let stream = rodio::microphone::MicrophoneBuilder::new()
168 .default_device()?
169 .default_config()?
170 .prefer_sample_rates([
171 SAMPLE_RATE, // sample rates trivially resamplable to `SAMPLE_RATE`
172 SAMPLE_RATE.saturating_mul(nz!(2)),
173 SAMPLE_RATE.saturating_mul(nz!(3)),
174 SAMPLE_RATE.saturating_mul(nz!(4)),
175 ])
176 .prefer_channel_counts([nz!(1), nz!(2), nz!(3), nz!(4)])
177 .prefer_buffer_sizes(512..)
178 .open_stream()?;
179 info!("Opened microphone: {:?}", stream.config());
180
181 let stream = stream
182 .possibly_disconnected_channels_to_mono()
183 .constant_samplerate(SAMPLE_RATE)
184 .limit(LimitSettings::live_performance())
185 .process_buffer::<BUFFER_SIZE, _>(move |buffer| {
186 let mut int_buffer: [i16; _] = buffer.map(|s| s.to_sample());
187 if voip_parts
188 .echo_canceller
189 .lock()
190 .process_stream(
191 &mut int_buffer,
192 SAMPLE_RATE.get() as i32,
193 CHANNEL_COUNT.get() as i32,
194 )
195 .context("livekit audio processor error")
196 .log_err()
197 .is_some()
198 {
199 for (sample, processed) in buffer.iter_mut().zip(&int_buffer) {
200 *sample = (*processed).to_sample();
201 }
202 }
203 })
204 .denoise()
205 .context("Could not set up denoiser")?
206 .automatic_gain_control(0.90, 1.0, 0.0, 5.0)
207 .periodic_access(Duration::from_millis(100), move |agc_source| {
208 agc_source
209 .set_enabled(LIVE_SETTINGS.auto_microphone_volume.load(Ordering::Relaxed));
210 let denoise = agc_source.inner_mut();
211 denoise.set_enabled(LIVE_SETTINGS.denoise.load(Ordering::Relaxed));
212 });
213
214 let stream = if voip_parts.legacy_audio_compatible {
215 stream.constant_params(LEGACY_CHANNEL_COUNT, LEGACY_SAMPLE_RATE)
216 } else {
217 stream.constant_params(CHANNEL_COUNT, SAMPLE_RATE)
218 };
219
220 let (replay, stream) = stream.replayable(REPLAY_DURATION)?;
221 voip_parts
222 .replays
223 .add_voip_stream("local microphone".to_string(), replay);
224
225 Ok(stream)
226 }
227
228 pub fn play_voip_stream(
229 source: impl rodio::Source + Send + 'static,
230 speaker_name: String,
231 is_staff: bool,
232 cx: &mut App,
233 ) -> anyhow::Result<()> {
234 let (replay_source, source) = source
235 .constant_params(CHANNEL_COUNT, SAMPLE_RATE)
236 .automatic_gain_control(0.90, 1.0, 0.0, 5.0)
237 .periodic_access(Duration::from_millis(100), move |agc_source| {
238 agc_source.set_enabled(LIVE_SETTINGS.auto_speaker_volume.load(Ordering::Relaxed));
239 })
240 .replayable(REPLAY_DURATION)
241 .expect("REPLAY_DURATION is longer than 100ms");
242
243 cx.update_default_global(|this: &mut Self, _cx| {
244 let output_mixer = this
245 .ensure_output_exists()
246 .context("Could not get output mixer")?;
247 output_mixer.add(source);
248 if is_staff {
249 this.replays.add_voip_stream(speaker_name, replay_source);
250 }
251 Ok(())
252 })
253 }
254
255 pub fn play_sound(sound: Sound, cx: &mut App) {
256 cx.update_default_global(|this: &mut Self, cx| {
257 let source = this.sound_source(sound, cx).log_err()?;
258 let output_mixer = this
259 .ensure_output_exists()
260 .context("Could not get output mixer")
261 .log_err()?;
262
263 output_mixer.add(source);
264 Some(())
265 });
266 }
267
268 pub fn end_call(cx: &mut App) {
269 cx.update_default_global(|this: &mut Self, _cx| {
270 this.output_handle.take();
271 });
272 }
273
274 fn sound_source(&mut self, sound: Sound, cx: &App) -> Result<impl Source + use<>> {
275 if let Some(wav) = self.source_cache.get(&sound) {
276 return Ok(wav.clone());
277 }
278
279 let path = format!("sounds/{}.wav", sound.file());
280 let bytes = cx
281 .asset_source()
282 .load(&path)?
283 .map(anyhow::Ok)
284 .with_context(|| format!("No asset available for path {path}"))??
285 .into_owned();
286 let cursor = Cursor::new(bytes);
287 let source = Decoder::new(cursor)?.buffered();
288
289 self.source_cache.insert(sound, source.clone());
290
291 Ok(source)
292 }
293}
294
295#[cfg(not(any(all(target_os = "windows", target_env = "gnu"), target_os = "freebsd")))]
296pub struct VoipParts {
297 echo_canceller: Arc<Mutex<apm::AudioProcessingModule>>,
298 replays: replays::Replays,
299 legacy_audio_compatible: bool,
300}
301
302#[cfg(not(any(all(target_os = "windows", target_env = "gnu"), target_os = "freebsd")))]
303impl VoipParts {
304 pub fn new(cx: &AsyncApp) -> anyhow::Result<Self> {
305 let (apm, replays) = cx.try_read_default_global::<Audio, _>(|audio, _| {
306 (Arc::clone(&audio.echo_canceller), audio.replays.clone())
307 })?;
308 let legacy_audio_compatible =
309 AudioSettings::try_read_global(cx, |settings| settings.legacy_audio_compatible)
310 .unwrap_or(true);
311
312 Ok(Self {
313 legacy_audio_compatible,
314 echo_canceller: apm,
315 replays,
316 })
317 }
318}