Skip to content

YC (S26) | Open Computer History | Continuously record your company computer work, map your workflows, help you find work worth automating, and power your agents' context

fix(audio): feed contiguous native frames to speech detection (#7365)

Louis Beaumont6e531145bc85
diff --git a/crates/screenpipe-audio/src/speaker/prepare_segments.rs b/crates/screenpipe-audio/src/speaker/prepare_segments.rs
index 185dd715a..56f1f4d9e 100644
--- a/crates/screenpipe-audio/src/speaker/prepare_segments.rs
+++ b/crates/screenpipe-audio/src/speaker/prepare_segments.rs
@@ -35,14 +35,9 @@ pub async fn prepare_segments(
filter_music_frames(&mut audio_data);
}
- // Silero VAD v5 expects continuous 512-sample chunks at 16kHz (32ms).
- // On Windows, WASAPI delivers lower audio levels than CoreAudio, so we
- // must feed Silero at its native frame size to preserve its LSTM temporal
- // state — using 1600 caused 68ms gaps that broke speech detection.
- #[cfg(target_os = "windows")]
- let frame_size = 512;
- #[cfg(not(target_os = "windows"))]
- let frame_size = 1600;
+ // Feed every sample to the selected engine at its native window size.
+ // Oversized Silero input is truncated internally, creating detection gaps.
+ let frame_size = vad_engine.lock().await.frame_size();
let vad_engine = vad_engine.clone();
// Use a lower speech threshold for output/system audio devices.
@@ -60,11 +55,20 @@ pub async fn prepare_segments(
let mut total_frames = 0;
let mut speech_frame_count = 0;
+ let mut final_frame = vec![0.0; frame_size];
for chunk in audio_data.chunks(frame_size) {
total_frames += 1;
let mut new_chunk = chunk.to_vec();
- let status = vad_engine.lock().await.audio_type(chunk);
+ // Pad only the final analysis window. Keep the original samples and
+ // duration for persistence/transcription, including a short final cue.
+ let vad_frame = if chunk.len() < frame_size {
+ final_frame[..chunk.len()].copy_from_slice(chunk);
+ &final_frame[..]
+ } else {
+ chunk
+ };
+ let status = vad_engine.lock().await.audio_type(vad_frame);
match status {
Ok(VadStatus::Speech) => {
if let Ok(processed_audio) = spectral_subtraction(chunk, noise) {
diff --git a/crates/screenpipe-audio/src/vad/mod.rs b/crates/screenpipe-audio/src/vad/mod.rs
index d387ee1ab..085c873ec 100644
--- a/crates/screenpipe-audio/src/vad/mod.rs
+++ b/crates/screenpipe-audio/src/vad/mod.rs
@@ -52,6 +52,12 @@ pub enum VadEngineEnum {
pub const OUTPUT_SPEECH_THRESHOLD: f32 = 0.15;
pub trait VadEngine: Send {
+ /// Native analysis window at the pipeline's 16 kHz sample rate.
+ /// Silero requires 512 contiguous samples; other engines may override it.
+ fn frame_size(&self) -> usize {
+ 512
+ }
+
fn is_voice_segment(&mut self, audio_chunk: &[f32]) -> anyhow::Result<bool>;
fn audio_type(&mut self, audio_chunk: &[f32]) -> anyhow::Result<VadStatus>;
/// Override the speech probability threshold. Call with `None` to reset to default.
diff --git a/crates/screenpipe-audio/src/vad/webrtc.rs b/crates/screenpipe-audio/src/vad/webrtc.rs
index c67fef19a..aaee13975 100644
--- a/crates/screenpipe-audio/src/vad/webrtc.rs
+++ b/crates/screenpipe-audio/src/vad/webrtc.rs
@@ -1,20 +1,32 @@
+// screenpipe — AI that knows everything you've seen, said, or heard
+// https://screenpipe.com
+
use anyhow;
use vad_rs::VadStatus;
use super::VadEngine;
-#[derive(Default)]
pub struct WebRtcVad {
vad: webrtc_vad::Vad,
}
impl WebRtcVad {
pub fn new() -> Self {
- let vad = webrtc_vad::Vad::new();
+ let vad = webrtc_vad::Vad::new_with_rate(webrtc_vad::SampleRate::Rate16kHz);
Self { vad }
}
}
+impl Default for WebRtcVad {
+ fn default() -> Self {
+ Self::new()
+ }
+}
impl VadEngine for WebRtcVad {
+ fn frame_size(&self) -> usize {
+ // WebRTC accepts 10, 20 or 30 ms frames, not Silero's 32 ms window.
+ 480
+ }
+
fn is_voice_segment(&mut self, audio_chunk: &[f32]) -> anyhow::Result<bool> {
// Convert f32 to i16
let i16_chunk: Vec<i16> = audio_chunk.iter().map(|&x| (x * 32767.0) as i16).collect();
diff --git a/crates/screenpipe-audio/tests/audio_pipeline_benchmark/cross_device_benchmark.rs b/crates/screenpipe-audio/tests/audio_pipeline_benchmark/cross_device_benchmark.rs
index a2fa7fa2b..9618626e4 100644
--- a/crates/screenpipe-audio/tests/audio_pipeline_benchmark/cross_device_benchmark.rs
+++ b/crates/screenpipe-audio/tests/audio_pipeline_benchmark/cross_device_benchmark.rs
@@ -17,15 +17,12 @@ use screenpipe_audio::vad::silero::SileroVad;
use screenpipe_audio::vad::VadEngine;
use vad_rs::VadStatus;
-/// Frame size matching prepare_segments.rs.
-const FRAME_SIZE: usize = 1600;
-
/// Compute speech_ratio for an audio segment.
fn compute_speech_ratio(audio: &[f32], vad: &mut SileroVad) -> f32 {
let mut total_frames = 0u32;
let mut speech_frames = 0u32;
- for chunk in audio.chunks(FRAME_SIZE) {
+ for chunk in audio.chunks(vad.frame_size()) {
total_frames += 1;
if let Ok(VadStatus::Speech) = vad.audio_type(chunk) {
speech_frames += 1;
diff --git a/crates/screenpipe-audio/tests/audio_pipeline_benchmark/pipeline_benchmark.rs b/crates/screenpipe-audio/tests/audio_pipeline_benchmark/pipeline_benchmark.rs
index d761e0ed3..caf65d22e 100644
--- a/crates/screenpipe-audio/tests/audio_pipeline_benchmark/pipeline_benchmark.rs
+++ b/crates/screenpipe-audio/tests/audio_pipeline_benchmark/pipeline_benchmark.rs
@@ -22,9 +22,6 @@ use vad_rs::VadStatus;
use std::sync::Arc;
-/// Frame size matching prepare_segments.rs.
-const FRAME_SIZE: usize = 1600;
-
/// Chunk duration in seconds (matching production pipeline).
const CHUNK_DURATION_SECS: f64 = 30.0;
@@ -59,7 +56,7 @@ fn simulate_pipeline(
let mut total_frames = 0u32;
let mut speech_frames = 0u32;
- for frame in normalized.chunks(FRAME_SIZE) {
+ for frame in normalized.chunks(vad.frame_size()) {
total_frames += 1;
if let Ok(VadStatus::Speech) = vad.audio_type(frame) {
speech_frames += 1;
diff --git a/crates/screenpipe-audio/tests/audio_pipeline_benchmark/vad_benchmark.rs b/crates/screenpipe-audio/tests/audio_pipeline_benchmark/vad_benchmark.rs
index 2245f1e85..efc919985 100644
--- a/crates/screenpipe-audio/tests/audio_pipeline_benchmark/vad_benchmark.rs
+++ b/crates/screenpipe-audio/tests/audio_pipeline_benchmark/vad_benchmark.rs
@@ -24,9 +24,6 @@ use vad_rs::VadStatus;
/// The thresholds to sweep. Includes the current production value (0.05).
const THRESHOLDS: &[f32] = &[0.005, 0.01, 0.02, 0.03, 0.04, 0.05, 0.10, 0.20];
-/// Frame size matching prepare_segments.rs (100ms at 16kHz).
-const FRAME_SIZE: usize = 1600;
-
/// Chunk duration matching the audio pipeline (30 seconds).
const CHUNK_DURATION_SECS: f64 = 30.0;
@@ -52,7 +49,7 @@ fn compute_speech_ratio(audio_chunk: &[f32], vad: &mut SileroVad) -> f32 {
let mut total_frames = 0u32;
let mut speech_frame_count = 0u32;
- for chunk in audio_chunk.chunks(FRAME_SIZE) {
+ for chunk in audio_chunk.chunks(vad.frame_size()) {
total_frames += 1;
let status = vad.audio_type(chunk);
if let Ok(VadStatus::Speech) = status {
diff --git a/crates/screenpipe-audio/tests/vad_frame_continuity_test.rs b/crates/screenpipe-audio/tests/vad_frame_continuity_test.rs
new file mode 100644
index 000000000..ade1af73d
--- /dev/null
+++ b/crates/screenpipe-audio/tests/vad_frame_continuity_test.rs
@@ -0,0 +1,186 @@
+// screenpipe — AI that knows everything you've seen, said, or heard
+// https://screenpipe.com
+
+use screenpipe_audio::speaker::{embedding_manager::EmbeddingManager, prepare_segments};
+use screenpipe_audio::vad::{webrtc::WebRtcVad, VadEngine};
+use std::sync::{Arc, Mutex};
+use vad_rs::VadStatus;
+
+// Model the native Silero window: samples beyond its first 512 never reach
+// inference. Put a short cue in the portion the old macOS frame loop skipped.
+struct WindowVad {
+ seen: Arc<Mutex<Vec<Vec<f32>>>>,
+}
+impl VadEngine for WindowVad {
+ fn is_voice_segment(&mut self, audio: &[f32]) -> anyhow::Result<bool> {
+ Ok(matches!(self.audio_type(audio)?, VadStatus::Speech))
+ }
+ fn audio_type(&mut self, audio: &[f32]) -> anyhow::Result<VadStatus> {
+ self.seen.lock().unwrap().push(audio.to_vec());
+ Ok(if audio.iter().take(512).any(|x| x.abs() > 0.1) {
+ VadStatus::Speech
+ } else {
+ VadStatus::Silence
+ })
+ }
+ fn set_speech_threshold(&mut self, _: Option<f32>) {}
+}
+
+#[tokio::test]
+async fn short_cue_between_native_windows_reaches_transcription() {
+ let seen = Arc::new(Mutex::new(Vec::new()));
+ let vad: Arc<tokio::sync::Mutex<Box<dyn VadEngine + Send>>> =
+ Arc::new(tokio::sync::Mutex::new(Box::new(WindowVad {
+ seen: seen.clone(),
+ })));
+ let mut audio = vec![0.0; 1601];
+ for (i, sample) in audio[600..1000].iter_mut().enumerate() {
+ *sample = if i % 2 == 0 { 0.5 } else { -0.5 };
+ }
+ let (mut segments, accepted, _) = prepare_segments(
+ &audio,
+ vad,
+ None,
+ Arc::new(Mutex::new(EmbeddingManager::new(4))),
+ None,
+ "synthetic microphone",
+ false,
+ false,
+ )
+ .await
+ .unwrap();
+ assert!(
+ accepted,
+ "a cue outside the first 512 samples must not be skipped"
+ );
+ let segment = segments
+ .recv()
+ .await
+ .expect("retain the full audio for transcription");
+ assert_eq!(segment.samples.len(), audio.len());
+ assert!(segments.recv().await.is_none());
+ let calls = seen.lock().unwrap();
+ assert_eq!(calls.len(), 4);
+ assert!(calls.iter().all(|call| call.len() == 512));
+}
+
+#[test]
+fn webrtc_accepts_30ms_of_16khz_audio() {
+ let mut vad = WebRtcVad::new();
+ assert!(
+ vad.audio_type(&[0.0; 480]).is_ok(),
+ "fallback must use the pipeline's 16 kHz sample rate"
+ );
+}
+
+struct CheckedWebRtc {
+ inner: WebRtcVad,
+ calls: Arc<Mutex<usize>>,
+}
+impl VadEngine for CheckedWebRtc {
+ fn frame_size(&self) -> usize {
+ self.inner.frame_size()
+ }
+ fn is_voice_segment(&mut self, audio: &[f32]) -> anyhow::Result<bool> {
+ self.inner.is_voice_segment(audio)
+ }
+ fn audio_type(&mut self, audio: &[f32]) -> anyhow::Result<VadStatus> {
+ assert_eq!(audio.len(), 480);
+ let result = self.inner.audio_type(audio);
+ assert!(
+ result.is_ok(),
+ "fallback must receive a supported frame, including the tail"
+ );
+ *self.calls.lock().unwrap() += 1;
+ result
+ }
+ fn set_speech_threshold(&mut self, value: Option<f32>) {
+ self.inner.set_speech_threshold(value)
+ }
+}
+
+#[tokio::test]
+async fn fallback_processes_a_partial_final_frame_without_invalid_length_errors() {
+ let calls = Arc::new(Mutex::new(0));
+ let vad: Arc<tokio::sync::Mutex<Box<dyn VadEngine + Send>>> =
+ Arc::new(tokio::sync::Mutex::new(Box::new(CheckedWebRtc {
+ inner: WebRtcVad::default(),
+ calls: calls.clone(),
+ })));
+ let (mut segments, accepted, ratio) = prepare_segments(
+ &[0.0; 1601],
+ vad,
+ None,
+ Arc::new(Mutex::new(EmbeddingManager::new(4))),
+ None,
+ "synthetic microphone",
+ false,
+ false,
+ )
+ .await
+ .unwrap();
+ assert!(!accepted);
+ assert_eq!(ratio, 0.0);
+ assert!(segments.recv().await.is_none());
+ assert_eq!(*calls.lock().unwrap(), 4);
+}
+
+/// Bounded native-model measurement on a public repository speech fixture.
+/// Set VAD_SPEECH_FIXTURE to the Git LFS contents of test_data/accuracy1.wav.
+/// Reports timing without a machine-dependent latency assertion.
+#[tokio::test]
+#[ignore = "requires the Silero model and a decoded Git LFS speech fixture"]
+async fn native_silero_speech_continuity_cost() {
+ use screenpipe_audio::vad::silero::SileroVad;
+ use std::time::Instant;
+ let path = std::env::var("VAD_SPEECH_FIXTURE").expect("set VAD_SPEECH_FIXTURE");
+ let (audio, rate) = screenpipe_audio::pcm_decode(std::path::Path::new(&path)).unwrap();
+ let mut audio = screenpipe_audio::resample(&audio, rate, 16000).unwrap();
+ audio.truncate(30 * 16000);
+ assert!(
+ audio.len() >= 16000,
+ "need at least one second of real audio"
+ );
+ let audio = screenpipe_audio::utils::audio::normalize_v2(&audio);
+ SileroVad::ensure_model_available().await.unwrap();
+ for frame_size in [1600, 512] {
+ let mut vad = SileroVad::new().await.unwrap();
+ let start = Instant::now();
+ let mut speech = 0;
+ for chunk in audio.chunks(frame_size) {
+ speech += usize::from(matches!(vad.audio_type(chunk).unwrap(), VadStatus::Speech));
+ }
+ println!(
+ "frame_size={frame_size} frames={} speech={speech} elapsed_ms={:.3}",
+ audio.chunks(frame_size).len(),
+ start.elapsed().as_secs_f64() * 1000.0
+ );
+ if frame_size == 512 {
+ assert!(speech > 0, "native detector must recognize fixture speech");
+ }
+ }
+ let vad: Arc<tokio::sync::Mutex<Box<dyn VadEngine + Send>>> = Arc::new(
+ tokio::sync::Mutex::new(Box::new(SileroVad::new().await.unwrap())),
+ );
+ let start = Instant::now();
+ let (mut segments, accepted, ratio) = prepare_segments(
+ &audio,
+ vad,
+ None,
+ Arc::new(Mutex::new(EmbeddingManager::new(4))),
+ None,
+ "public speech fixture",
+ false,
+ false,
+ )
+ .await
+ .unwrap();
+ println!(
+ "prepare_segments audio_seconds={:.3} elapsed_ms={:.3} speech_ratio={ratio}",
+ audio.len() as f64 / 16000.0,
+ start.elapsed().as_secs_f64() * 1000.0
+ );
+ assert!(accepted);
+ let segment = segments.recv().await.unwrap();
+ assert_eq!(segment.samples.len(), audio.len());
+}
diff --git a/docs/COVERAGE.md b/docs/COVERAGE.md
index 3b8b84d6e..76e86668e 100644
--- a/docs/COVERAGE.md
+++ b/docs/COVERAGE.md
@@ -37,16 +37,16 @@ results and `cargo llvm-cov` data on top when judging release confidence.
### Core Engine
- Mapped suites: 36
-- Mapped Rust files: 380
-- Active test blocks: 3735
-- Ignored/manual test blocks: 168
-- Weighted coverage points: 3081.1
+- Mapped Rust files: 381
+- Active test blocks: 3738
+- Ignored/manual test blocks: 169
+- Weighted coverage points: 3082.6
| Platform | Suites | Active tests | Ignored tests | Weighted points | Layers | Flows | Critical score |
| --- | --- | --- | --- | --- | --- | --- | --- |
-| windows | 32 | 3589 | 154 | 3014.9 | 21 | 11 | 100% |
-| macos | 31 | 3617 | 129 | 3004.8 | 22 | 11 | 100% |
-| linux | 28 | 3174 | 129 | 2639.6 | 20 | 11 | 100% |
+| windows | 32 | 3592 | 155 | 3016.4 | 21 | 11 | 100% |
+| macos | 31 | 3620 | 130 | 3006.3 | 22 | 11 | 100% |
+| linux | 28 | 3177 | 130 | 2641.1 | 20 | 11 | 100% |
## Refresh
diff --git a/docs/coverage/CORE.md b/docs/coverage/CORE.md
index c6e0cd4db..a48225d9b 100644
--- a/docs/coverage/CORE.md
+++ b/docs/coverage/CORE.md
@@ -8,11 +8,11 @@ confidence, and criticality.
- Manifest: `docs/coverage/core-engine-map.json`
- Tracked crates: screenpipe-engine, screenpipe-db, screenpipe-sqlite-coordinator, screenpipe-audio, screenpipe-screen, screenpipe-a11y, screenpipe-fs
- Mapped suites: 36
-- Mapped Rust files: 380
-- Active test blocks: 3735
-- Ignored/manual test blocks: 168
-- Declared test blocks: 3903
-- Weighted coverage points: 3081.1
+- Mapped Rust files: 381
+- Active test blocks: 3738
+- Ignored/manual test blocks: 169
+- Declared test blocks: 3907
+- Weighted coverage points: 3082.6
Confidence weights: strong=1.0, partial=0.7, conditional=0.4, smoke=0.3.
Criticality weights: high=1.0, medium=0.7, low=0.4.
@@ -23,9 +23,9 @@ are explicitly enabled in a runtime lane.
| Platform | Suites | Active tests | Ignored tests | Weighted points | Layers | Flows | Critical score |
| --- | --- | --- | --- | --- | --- | --- | --- |
-| windows | 32 | 3589 | 154 | 3014.9 | 21 | 11 | 100% |
-| macos | 31 | 3617 | 129 | 3004.8 | 22 | 11 | 100% |
-| linux | 28 | 3174 | 129 | 2639.6 | 20 | 11 | 100% |
+| windows | 32 | 3592 | 155 | 3016.4 | 21 | 11 | 100% |
+| macos | 31 | 3620 | 130 | 3006.3 | 22 | 11 | 100% |
+| linux | 28 | 3177 | 130 | 2641.1 | 20 | 11 | 100% |
## Crate Summary
@@ -34,7 +34,7 @@ are explicitly enabled in a runtime lane.
| screenpipe-engine | 12 | 21 | 126 | 1796 | 52 | 1381.6 | 10 |
| screenpipe-db | 5 | 64 | 26 | 589 | 25 | 561.1 | 9 |
| screenpipe-sqlite-coordinator | 1 | 0 | 3 | 27 | 0 | 27.0 | 2 |
-| screenpipe-audio | 7 | 25 | 53 | 672 | 48 | 588.8 | 5 |
+| screenpipe-audio | 7 | 26 | 53 | 675 | 49 | 590.3 | 5 |
| screenpipe-screen | 6 | 9 | 19 | 272 | 9 | 242.3 | 4 |
| screenpipe-a11y | 4 | 2 | 31 | 376 | 34 | 277.3 | 3 |
| screenpipe-fs | 1 | 0 | 1 | 3 | 0 | 3.0 | 3 |
@@ -63,7 +63,7 @@ bun run coverage:core -- --llvm-cov-summary ../../docs/coverage/core-llvm-cov-su
| Layer | windows | macos | linux |
| --- | --- | --- | --- |
| accessibility | 4 suites / 374 active / 35 ignored / 340.1 pts | 4 suites / 415 active / 10 ignored / 336.4 pts | 4 suites / 334 active / 7 ignored / 306.5 pts |
-| audio | 9 suites / 794 active / 49 ignored / 710.8 pts | 8 suites / 788 active / 49 ignored / 706.6 pts | 7 suites / 709 active / 48 ignored / 651.3 pts |
+| audio | 9 suites / 797 active / 50 ignored / 712.3 pts | 8 suites / 791 active / 50 ignored / 708.0 pts | 7 suites / 712 active / 49 ignored / 652.8 pts |
| audio-device | 2 suites / 234 active / 7 ignored / 210.3 pts | 2 suites / 234 active / 7 ignored / 210.3 pts | 1 suites / 155 active / 6 ignored / 155.0 pts |
| configuration | 2 suites / 143 active / 3 ignored / 129.2 pts | 2 suites / 143 active / 3 ignored / 129.2 pts | 2 suites / 143 active / 3 ignored / 129.2 pts |
| database | 7 suites / 542 active / 21 ignored / 514.1 pts | 7 suites / 542 active / 21 ignored / 514.1 pts | 7 suites / 542 active / 21 ignored / 514.1 pts |
@@ -73,7 +73,7 @@ bun run coverage:core -- --llvm-cov-summary ../../docs/coverage/core-llvm-cov-su
| meeting | 6 suites / 1693 active / 24 ignored / 1382.5 pts | 6 suites / 1693 active / 24 ignored / 1382.5 pts | 4 suites / 1344 active / 20 ignored / 1057.2 pts |
| ocr | 4 suites / 137 active / 7 ignored / 127.7 pts | 4 suites / 130 active / 7 ignored / 125.5 pts | 3 suites / 121 active / 6 ignored / 116.5 pts |
| os-integration | 1 suites / 6 active / 0 ignored / 1.7 pts | 1 suites / 6 active / 0 ignored / 1.7 pts | - |
-| performance | 13 suites / 1622 active / 78 ignored / 1435.0 pts | 14 suites / 1730 active / 82 ignored / 1478.2 pts | 13 suites / 1622 active / 78 ignored / 1435.0 pts |
+| performance | 13 suites / 1625 active / 79 ignored / 1436.5 pts | 14 suites / 1733 active / 83 ignored / 1479.7 pts | 13 suites / 1625 active / 79 ignored / 1436.5 pts |
| pipes | 1 suites / 527 active / 3 ignored / 368.9 pts | 1 suites / 527 active / 3 ignored / 368.9 pts | 1 suites / 527 active / 3 ignored / 368.9 pts |
| privacy | 5 suites / 980 active / 42 ignored / 791.6 pts | 5 suites / 1021 active / 17 ignored / 787.9 pts | 5 suites / 940 active / 14 ignored / 757.9 pts |
| real-app | - | 1 suites / 108 active / 4 ignored / 43.2 pts | - |
@@ -81,7 +81,7 @@ bun run coverage:core -- --llvm-cov-summary ../../docs/coverage/core-llvm-cov-su
| storage | 4 suites / 667 active / 38 ignored / 553.9 pts | 4 suites / 667 active / 38 ignored / 553.9 pts | 4 suites / 667 active / 38 ignored / 553.9 pts |
| sync | 1 suites / 527 active / 3 ignored / 368.9 pts | 1 suites / 527 active / 3 ignored / 368.9 pts | 1 suites / 527 active / 3 ignored / 368.9 pts |
| timeline | 4 suites / 1214 active / 44 ignored / 994.4 pts | 4 suites / 1214 active / 44 ignored / 994.4 pts | 4 suites / 1214 active / 44 ignored / 994.4 pts |
-| transcription | 6 suites / 867 active / 47 ignored / 678.8 pts | 5 suites / 861 active / 47 ignored / 674.6 pts | 5 suites / 861 active / 47 ignored / 674.6 pts |
+| transcription | 6 suites / 870 active / 48 ignored / 680.2 pts | 5 suites / 864 active / 48 ignored / 676.0 pts | 5 suites / 864 active / 48 ignored / 676.0 pts |
| ui-events | 4 suites / 795 active / 34 ignored / 603.0 pts | 3 suites / 728 active / 5 ignored / 556.1 pts | 3 suites / 728 active / 5 ignored / 556.1 pts |
| vision-capture | 6 suites / 569 active / 32 ignored / 448.4 pts | 6 suites / 562 active / 32 ignored / 446.2 pts | 5 suites / 553 active / 31 ignored / 437.2 pts |
@@ -126,7 +126,7 @@ bun run coverage:core -- --llvm-cov-summary ../../docs/coverage/core-llvm-cov-su
| audio-device-stream-health | screenpipe-audio | windows, macos, linux | audio-device, audio, performance | audio-device-health, audio-record-transcribe, performance-liveness | high | strong | mixed | 14 | 155 | 6 | Device monitor, device manager, stream buffering, source lag, audio metrics, Bluetooth gap/hallucination regressions, and cross-platform process-tap watchdog counters (process_tap.rs split into src/core/process_tap/ modules). |
| audio-meetings-speakers-dedup | screenpipe-audio | windows, macos, linux | audio, meeting, speaker, transcription | audio-record-transcribe, meeting-live-notes, performance-liveness | high | strong | mixed | 26 | 269 | 10 | Meeting streaming config/controller logic, speaker embedding state, cross-device dedupe simulations, and overlap cleanup coverage. |
| audio-models-filtering | screenpipe-audio | windows, macos, linux | audio, transcription, privacy | audio-record-transcribe, privacy-and-redaction | medium | partial | mixed | 6 | 20 | 10 | Model-download/TLS guards, ONNX startup smoke, and music-versus-speech filtering. |
-| audio-pipeline-benchmarks | screenpipe-audio | windows, macos, linux | audio, transcription, performance | audio-record-transcribe, meeting-live-notes, performance-liveness | medium | partial | benchmark | 8 | 22 | 12 | Benchmark-backed regression probes for VAD, smart mode, meeting audio, quality, cross-device, and end-to-end pipeline timing. |
+| audio-pipeline-benchmarks | screenpipe-audio | windows, macos, linux | audio, transcription, performance | audio-record-transcribe, meeting-live-notes, performance-liveness | medium | partial | benchmark | 9 | 25 | 13 | Benchmark-backed regression probes for VAD, smart mode, meeting audio, quality, cross-device, and end-to-end pipeline timing. Native-window continuity, valid 16 kHz fallback frames and partial-tail regression tests; optional real-model speech-fixture timing is ignored by default. |
| audio-platform-output-capture | screenpipe-audio | windows, macos | audio-device, audio, meeting | audio-device-health, audio-record-transcribe, meeting-live-notes | high | partial | unit | 7 | 79 | 1 | OS-specific output/system-audio capture: CoreAudio process tap and SCK output watchdog plus VPIO health policy on macOS, per-process meeting audio taps on both platforms, and the Windows follow-the-audio output watchdog. Platform impl files are cfg-gated to their target OS. |
| audio-transcription-pipeline | screenpipe-audio | windows, macos, linux | audio, transcription, performance | audio-record-transcribe, meeting-live-notes, performance-liveness | high | partial | mixed | 16 | 121 | 9 | Batch deferral, cleanup, language detection, result normalization, and real recording/transcription tests. Hardware/model-heavy tests are ignored by default. Also covers ffmpeg encode child reaping on error paths. |
| audio-windows-parakeet-provider | screenpipe-audio | windows | audio, transcription | audio-record-transcribe | high | partial | unit | 1 | 6 | 0 | Registers existing CPU override, eligible adapter selection, DXGI ordinal, error fallback and adapter determinism tests. Windows-only; manifest registration does not verify execution on physical GPU hardware. |
@@ -271,6 +271,7 @@ bun run coverage:core -- --llvm-cov-summary ../../docs/coverage/core-llvm-cov-su
| audio-meetings-speakers-dedup | screenpipe-audio | tests/speaker_identification.rs | integration | 2 | 1 | 3 |
| audio-meetings-speakers-dedup | screenpipe-audio | tests/speaker_identity_gate_real_audio_test.rs | integration | 0 | 4 | 4 |
| audio-meetings-speakers-dedup | screenpipe-audio | tests/speaker_identity_gate_test.rs | integration | 11 | 0 | 11 |
+| audio-pipeline-benchmarks | screenpipe-audio | tests/vad_frame_continuity_test.rs | integration | 3 | 1 | 4 |
| db-runtime-reliability | screenpipe-db | src/cancellable_query.rs | source | 2 | 0 | 2 |
| db-timeline-frames | screenpipe-db | src/db/activity_ledger.rs | source | 5 | 0 | 5 |
| db-accessibility-ui-events | screenpipe-db | src/db/elements.rs | source | 5 | 0 | 5 |
diff --git a/docs/coverage/core-engine-map.json b/docs/coverage/core-engine-map.json
index 6692c0ca4..8d613b0af 100644
--- a/docs/coverage/core-engine-map.json
+++ b/docs/coverage/core-engine-map.json
@@ -1050,7 +1050,8 @@
"tests/audio_pipeline_benchmark/pipeline_benchmark.rs",
"tests/audio_pipeline_benchmark/quality_regression.rs",
"tests/audio_pipeline_benchmark/smart_mode_benchmark.rs",
- "tests/audio_pipeline_benchmark/vad_benchmark.rs"
+ "tests/audio_pipeline_benchmark/vad_benchmark.rs",
+ "tests/vad_frame_continuity_test.rs"
],
"platforms": [
"windows",
@@ -1070,7 +1071,7 @@
"criticality": "medium",
"confidence": "partial",
"kind": "benchmark",
- "notes": "Benchmark-backed regression probes for VAD, smart mode, meeting audio, quality, cross-device, and end-to-end pipeline timing."
+ "notes": "Benchmark-backed regression probes for VAD, smart mode, meeting audio, quality, cross-device, and end-to-end pipeline timing. Native-window continuity, valid 16 kHz fallback frames and partial-tail regression tests; optional real-model speech-fixture timing is ignored by default."
},
{
"id": "audio-models-filtering",