extract-speech 0.9.1

Voice activity detection inference for Silero, PulseVAD, PyAnnote, FSMN, TEN VAD, and MarbleNet
Documentation
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
//! Generic probability-to-speech-segment iterator.

// Duration thresholds are expressed in seconds and converted from bounded sample counts.
#![allow(clippy::cast_precision_loss)]

use log::debug;

use crate::{utils, SAMPLE_RATE};

pub trait VadModel {
    /// Resets model state before processing an independent audio stream.
    ///
    /// # Errors
    ///
    /// Returns a backend-specific reset error.
    fn reset(&mut self) -> anyhow::Result<()>;

    /// Computes the speech probability for one audio frame.
    ///
    /// # Errors
    ///
    /// Returns a backend-specific inference or input-validation error.
    fn probability(&mut self, audio_frame: &[f32]) -> anyhow::Result<f32>;
}

#[derive(Debug)]
pub struct VadIter<M> {
    model: M,
    params: utils::VadParams,
    state: State,
}

impl<M: VadModel> VadIter<M> {
    #[must_use]
    pub fn new(model: M, params: utils::VadParams) -> Self {
        debug!("vad_params: {params:?}");

        Self {
            model,
            params,
            state: State::default(),
        }
    }

    /// Detects speech segments in one complete waveform.
    ///
    /// # Errors
    ///
    /// Returns an error if parameters or samples are invalid, arithmetic
    /// overflows, or the model cannot be reset or evaluated.
    pub fn process(&mut self, samples: &[f32]) -> anyhow::Result<&[utils::TimeStamp]> {
        let params = Params::try_from(self.params.clone())?;
        validate_samples(samples)?;
        self.reset_states()?;

        for audio_frame in samples.chunks_exact(params.frame_size_samples) {
            let speech_probability = self.model.probability(audio_frame)?;
            self.state.update(&params, speech_probability)?;
        }

        self.state.finish(samples.len(), &params);
        apply_speech_padding(
            &mut self.state.speeches,
            params.speech_pad_samples,
            samples.len(),
        );

        Ok(&self.state.speeches)
    }

    fn reset_states(&mut self) -> anyhow::Result<()> {
        self.state = State::default();
        self.model.reset()
    }
}

#[cfg(any(feature = "candle", feature = "onnxruntime"))]
pub(crate) fn segment_probabilities(
    probabilities: &[f32],
    total_samples: usize,
    params: &utils::VadParams,
) -> anyhow::Result<Vec<utils::TimeStamp>> {
    let frame_size_samples = params
        .frame_size
        .checked_mul(params.sample_rate / 1_000)
        .ok_or_else(|| anyhow::anyhow!("VAD frame size exceeds usize"))?;
    segment_probabilities_with_frame_size(probabilities, total_samples, params, frame_size_samples)
}

#[cfg(any(feature = "candle", feature = "onnxruntime"))]
pub(crate) fn segment_probabilities_with_frame_size(
    probabilities: &[f32],
    total_samples: usize,
    params: &utils::VadParams,
    frame_size_samples: usize,
) -> anyhow::Result<Vec<utils::TimeStamp>> {
    let params = Params::try_from_with_frame_size(params, frame_size_samples)?;
    let mut state = State::default();
    for &probability in probabilities {
        state.update(&params, probability)?;
    }
    state.finish(total_samples, &params);
    apply_speech_padding(
        &mut state.speeches,
        params.speech_pad_samples,
        total_samples,
    );
    Ok(state.speeches)
}

#[derive(Debug)]
struct Params {
    threshold: f32,
    sample_rate: usize,
    frame_size_samples: usize,
    min_speech_samples: usize,
    speech_pad_samples: usize,
    max_speech_samples: Option<usize>,
    min_silence_samples: usize,
    min_silence_samples_at_max_speech: usize,
}

impl TryFrom<utils::VadParams> for Params {
    type Error = anyhow::Error;

    fn try_from(value: utils::VadParams) -> anyhow::Result<Self> {
        validate_parameters(&value)?;
        let samples_per_ms = value.sample_rate / 1000;
        let frame_size_samples = value
            .frame_size
            .checked_mul(samples_per_ms)
            .ok_or_else(|| anyhow::anyhow!("VAD frame size exceeds usize"))?;
        Self::try_from_with_frame_size(&value, frame_size_samples)
    }
}

impl Params {
    fn try_from_with_frame_size(
        value: &utils::VadParams,
        frame_size_samples: usize,
    ) -> anyhow::Result<Self> {
        validate_parameters(value)?;
        anyhow::ensure!(
            frame_size_samples > 0,
            "VAD frame size must be greater than zero"
        );

        let samples_per_ms = value.sample_rate / 1_000;
        let speech_pad_samples = samples_per_ms
            .checked_mul(value.speech_pad_ms)
            .ok_or_else(|| anyhow::anyhow!("speech padding exceeds usize"))?;
        let min_speech_samples = samples_per_ms
            .checked_mul(value.min_speech_duration_ms)
            .ok_or_else(|| anyhow::anyhow!("minimum speech duration exceeds usize"))?;
        let min_silence_samples = samples_per_ms
            .checked_mul(value.min_silence_duration_ms)
            .ok_or_else(|| anyhow::anyhow!("minimum silence duration exceeds usize"))?;
        let min_silence_samples_at_max_speech = samples_per_ms
            .checked_mul(98)
            .ok_or_else(|| anyhow::anyhow!("maximum-speech silence duration exceeds usize"))?;

        let max_speech_samples = if value.max_speech_duration_s.is_infinite() {
            None
        } else {
            let total = seconds_to_samples(value.max_speech_duration_s, value.sample_rate)?;
            let reserved = speech_pad_samples
                .checked_mul(2)
                .and_then(|padding| frame_size_samples.checked_add(padding))
                .ok_or_else(|| anyhow::anyhow!("maximum speech duration adjustment overflowed"))?;
            anyhow::ensure!(
                total > reserved,
                "maximum speech duration must exceed one frame plus twice the speech padding"
            );
            Some(total - reserved)
        };

        Ok(Self {
            threshold: value.threshold,
            sample_rate: value.sample_rate,
            frame_size_samples,
            min_speech_samples,
            speech_pad_samples,
            max_speech_samples,
            min_silence_samples,
            min_silence_samples_at_max_speech,
        })
    }
}

pub(crate) fn validate_parameters(params: &utils::VadParams) -> anyhow::Result<()> {
    anyhow::ensure!(
        params.sample_rate == SAMPLE_RATE,
        "VAD inference requires {SAMPLE_RATE} Hz samples"
    );
    anyhow::ensure!(
        params.frame_size > 0,
        "frame size must be greater than zero"
    );
    anyhow::ensure!(
        params.threshold.is_finite() && (0.0..=1.0).contains(&params.threshold),
        "threshold must be between 0 and 1"
    );
    anyhow::ensure!(
        params.max_speech_duration_s > 0.0,
        "maximum speech duration must be greater than zero"
    );

    let samples_per_ms = params.sample_rate / 1_000;
    let frame_size_samples = params
        .frame_size
        .checked_mul(samples_per_ms)
        .ok_or_else(|| anyhow::anyhow!("VAD frame size exceeds usize"))?;
    params
        .min_speech_duration_ms
        .checked_mul(samples_per_ms)
        .ok_or_else(|| anyhow::anyhow!("minimum speech duration exceeds usize"))?;
    params
        .min_silence_duration_ms
        .checked_mul(samples_per_ms)
        .ok_or_else(|| anyhow::anyhow!("minimum silence duration exceeds usize"))?;
    let speech_pad_samples = params
        .speech_pad_ms
        .checked_mul(samples_per_ms)
        .and_then(|samples| samples.checked_mul(2))
        .ok_or_else(|| anyhow::anyhow!("speech padding exceeds usize"))?;
    if params.max_speech_duration_s.is_finite() {
        let total = seconds_to_samples(params.max_speech_duration_s, params.sample_rate)?;
        let reserved = frame_size_samples
            .checked_add(speech_pad_samples)
            .ok_or_else(|| anyhow::anyhow!("maximum speech duration adjustment overflowed"))?;
        anyhow::ensure!(
            total > reserved,
            "maximum speech duration must exceed one frame plus twice the speech padding"
        );
    }
    Ok(())
}

pub(crate) fn validate_samples(samples: &[f32]) -> anyhow::Result<()> {
    anyhow::ensure!(
        samples
            .iter()
            .all(|sample| sample.is_finite() && (-1.0..=1.0).contains(sample)),
        "VAD samples must be finite and normalized to -1.0..=1.0"
    );
    Ok(())
}

#[allow(clippy::cast_possible_truncation, clippy::cast_sign_loss)] // The value is finite, positive, and range-checked before conversion.
pub(crate) fn seconds_to_samples(seconds: f32, sample_rate: usize) -> anyhow::Result<usize> {
    let samples = f64::from(seconds) * sample_rate as f64;
    anyhow::ensure!(samples.is_finite(), "sample duration must be finite");
    anyhow::ensure!(
        samples <= usize::MAX as f64,
        "sample duration exceeds usize"
    );
    Ok(samples as usize)
}

#[cfg(any(feature = "candle", feature = "onnxruntime"))]
pub(crate) fn milliseconds_to_samples(
    milliseconds: usize,
    sample_rate: usize,
    label: &str,
) -> anyhow::Result<usize> {
    milliseconds
        .checked_mul(sample_rate)
        .map(|samples| samples / 1_000)
        .ok_or_else(|| anyhow::anyhow!("{label} exceeds usize"))
}

#[derive(Debug, Default)]
struct State {
    current_sample: usize,
    temp_end: usize,
    next_start: usize,
    prev_end: usize,
    triggered: bool,
    current_speech: utils::TimeStamp,
    speeches: Vec<utils::TimeStamp>,
}

impl State {
    fn update(&mut self, params: &Params, speech_probability: f32) -> anyhow::Result<()> {
        anyhow::ensure!(
            speech_probability.is_finite(),
            "VAD model returned a non-finite speech probability"
        );
        self.current_sample = self
            .current_sample
            .checked_add(params.frame_size_samples)
            .ok_or_else(|| anyhow::anyhow!("VAD sample position overflowed"))?;

        let is_speech = speech_probability > params.threshold;
        if is_speech {
            if self.temp_end != 0 {
                self.temp_end = 0;
                if self.next_start < self.prev_end {
                    self.next_start = self
                        .current_sample
                        .saturating_sub(params.frame_size_samples);
                }
            }

            if !self.triggered {
                self.log_transition(speech_probability, params, "start");
                self.triggered = true;
                self.current_speech.start = self
                    .current_sample
                    .saturating_sub(params.frame_size_samples);
            }
        }

        if self.triggered
            && params.max_speech_samples.is_some_and(|maximum| {
                self.current_sample
                    .saturating_sub(self.current_speech.start)
                    > maximum
            })
        {
            if self.prev_end > 0 {
                self.current_speech.end = self.prev_end;
                self.take_speech();
                if self.next_start < self.prev_end {
                    self.triggered = false;
                } else {
                    self.current_speech.start = self.next_start;
                }
            } else {
                self.current_speech.end = self.current_sample;
                self.take_speech();
                self.triggered = false;
            }
            self.clear_temporary_boundaries();
            return Ok(());
        }

        if is_speech {
            return Ok(());
        }

        if speech_probability >= params.threshold - 0.15 && speech_probability < params.threshold {
            let state = if self.triggered {
                "speaking"
            } else {
                "silence"
            };
            self.log_transition(speech_probability, params, state);
        }

        if self.triggered && speech_probability < params.threshold - 0.15 {
            self.log_transition(speech_probability, params, "end");
            if self.temp_end == 0 {
                self.temp_end = self.current_sample;
            }
            if self.current_sample.saturating_sub(self.temp_end)
                > params.min_silence_samples_at_max_speech
            {
                self.prev_end = self.temp_end;
            }
            if self.current_sample.saturating_sub(self.temp_end) >= params.min_silence_samples {
                self.finish_current_speech(self.temp_end, params.min_speech_samples);
            }
        }
        Ok(())
    }

    fn finish(&mut self, last_sample: usize, params: &Params) {
        if self.triggered {
            self.finish_current_speech(last_sample, params.min_speech_samples);
        }
    }

    fn finish_current_speech(&mut self, end: usize, min_speech_samples: usize) {
        self.current_speech.end = end;
        if self
            .current_speech
            .end
            .saturating_sub(self.current_speech.start)
            >= min_speech_samples
        {
            self.take_speech();
        } else {
            self.current_speech = utils::TimeStamp::default();
        }
        self.triggered = false;
        self.clear_temporary_boundaries();
    }

    fn take_speech(&mut self) {
        self.speeches.push(std::mem::take(&mut self.current_speech));
    }

    fn clear_temporary_boundaries(&mut self) {
        self.prev_end = 0;
        self.next_start = 0;
        self.temp_end = 0;
    }

    fn log_transition(&self, speech_probability: f32, params: &Params, title: &str) {
        let sample = self
            .current_sample
            .saturating_sub(params.frame_size_samples);
        debug!(
            "[{:10}: {:.3} s ({:.3}) {:8}]",
            title,
            sample as f32 / params.sample_rate as f32,
            speech_probability,
            sample,
        );
    }
}

pub(crate) fn apply_speech_padding(
    speeches: &mut [utils::TimeStamp],
    speech_pad_samples: usize,
    total_samples: usize,
) {
    if speeches.is_empty() {
        return;
    }

    speeches[0].start = speeches[0].start.saturating_sub(speech_pad_samples);

    for index in 0..speeches.len().saturating_sub(1) {
        let (left, right) = speeches.split_at_mut(index + 1);
        let current = &mut left[index];
        let next = &mut right[0];
        let silence = next.start.saturating_sub(current.end);

        if silence < speech_pad_samples.saturating_mul(2) {
            let half_silence = silence / 2;
            current.end = current.end.saturating_add(half_silence).min(total_samples);
            next.start = next.start.saturating_sub(silence - half_silence);
        } else {
            current.end = current
                .end
                .saturating_add(speech_pad_samples)
                .min(total_samples);
            next.start = next.start.saturating_sub(speech_pad_samples);
        }
    }

    if let Some(last) = speeches.last_mut() {
        last.end = last
            .end
            .saturating_add(speech_pad_samples)
            .min(total_samples);
    }
}

#[cfg(test)]
mod tests {
    use super::*;

    fn params() -> Params {
        Params::try_from(utils::VadParams::default()).unwrap()
    }

    #[test]
    fn trailing_speech_that_starts_at_zero_is_kept() {
        let params = params();
        let mut state = State::default();
        for _ in 0..10 {
            state.update(&params, 1.0).unwrap();
        }

        state.finish(state.current_sample, &params);

        assert_eq!(state.speeches.len(), 1);
        assert_eq!(state.speeches[0].start, 0);
        assert_eq!(state.speeches[0].end, 10 * params.frame_size_samples);
    }

    #[test]
    fn short_speech_is_discarded_and_state_is_reset() {
        let params = params();
        let mut state = State::default();
        state.update(&params, 1.0).unwrap();
        for _ in 0..5 {
            state.update(&params, 0.0).unwrap();
        }

        assert!(!state.triggered);
        assert!(state.speeches.is_empty());
        assert_eq!(state.current_speech, utils::TimeStamp::default());
    }

    #[test]
    fn padding_is_shared_between_close_segments() {
        let mut speeches = vec![
            utils::TimeStamp {
                start: 100,
                end: 200,
            },
            utils::TimeStamp {
                start: 220,
                end: 300,
            },
        ];

        apply_speech_padding(&mut speeches, 30, 400);

        assert_eq!(
            speeches[0],
            utils::TimeStamp {
                start: 70,
                end: 210
            }
        );
        assert_eq!(
            speeches[1],
            utils::TimeStamp {
                start: 210,
                end: 330
            }
        );
    }

    #[test]
    fn continuous_speech_is_split_at_the_maximum_duration() {
        let mut params = Params::try_from(utils::VadParams {
            min_speech_duration_ms: 0,
            speech_pad_ms: 0,
            max_speech_duration_s: 0.1,
            ..utils::VadParams::default()
        })
        .unwrap();
        params.min_silence_samples_at_max_speech = 0;
        let mut state = State::default();

        for _ in 0..6 {
            state.update(&params, 1.0).unwrap();
        }
        state.finish(state.current_sample, &params);

        assert_eq!(state.speeches.len(), 2);
        assert!(state.speeches.iter().all(|speech| {
            speech.end - speech.start <= seconds_to_samples(0.1, SAMPLE_RATE).unwrap()
        }));
    }

    struct ConstantModel;

    impl VadModel for ConstantModel {
        fn reset(&mut self) -> anyhow::Result<()> {
            Ok(())
        }

        fn probability(&mut self, _audio_frame: &[f32]) -> anyhow::Result<f32> {
            Ok(1.0)
        }
    }

    #[test]
    fn public_iterator_rejects_a_zero_frame_without_panicking() {
        let mut iterator = VadIter::new(
            ConstantModel,
            utils::VadParams {
                frame_size: 0,
                ..utils::VadParams::default()
            },
        );

        assert!(iterator.process(&[0.0; 512]).is_err());
    }

    #[test]
    fn parameter_validation_rejects_sample_count_overflow() {
        let params = utils::VadParams {
            speech_pad_ms: usize::MAX,
            ..utils::VadParams::default()
        };

        assert!(validate_parameters(&params).is_err());
    }
}