aria-kit 0.7.0

Audio capture, transcription, synthesis, and voice-analysis contracts for local speech apps
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

//! Voice-activity inference over canonical audio frames: which stretches
//! carry a human voice rather than music, noise, or silence.

mod endpointer;
#[cfg(all(feature = "silero", not(target_family = "wasm")))]
mod map;
#[cfg(feature = "silero")]
mod silero;

pub use endpointer::{Boundary, Endpointer};
#[cfg(all(feature = "silero", not(target_family = "wasm")))]
pub(crate) use map::Progress;
#[cfg(all(feature = "silero", not(target_family = "wasm")))]
pub use map::{SpeechDetector, SpeechMap};
#[cfg(feature = "silero")]
pub use silero::{SileroVad, Speech};

pub const SPEECH_THRESHOLD: f32 = 0.5;

/// The silence that closes an utterance in a [`SpeechMap`].
#[cfg(all(feature = "silero", not(target_family = "wasm")))]
pub(crate) const UTTERANCE_GAP: std::time::Duration = std::time::Duration::from_millis(700);

/// Frames to keep before a [`Boundary::Start`], so an utterance does not open
/// mid-syllable. `StopOnSilence` keeps this many.
pub const LEAD_IN_FRAMES: usize = 4;

pub trait VoiceActivityDetector {
    /// The exact number of canonical samples `probability` accepts per call.
    /// At least one.
    const FRAME_SAMPLES: usize;

    /// Clears state carried across recordings; stateless detectors keep the no-op.
    fn reset(&mut self) {}

    fn probability(&mut self, samples: &[i16]) -> Result<f32, VoiceActivityError>;
}

impl<V: VoiceActivityDetector> VoiceActivityDetector for &mut V {
    const FRAME_SAMPLES: usize = V::FRAME_SAMPLES;

    fn reset(&mut self) {
        (**self).reset();
    }

    fn probability(&mut self, samples: &[i16]) -> Result<f32, VoiceActivityError> {
        (**self).probability(samples)
    }
}

#[derive(Debug, thiserror::Error)]
pub enum VoiceActivityError {
    #[error("failed to load the bundled voice-activity model")]
    LoadModel(#[source] Box<dyn std::error::Error + Send + Sync>),

    #[error("voice-activity frame contains {actual_samples} samples; expected {expected_samples}")]
    InvalidFrameLength {
        actual_samples: usize,
        expected_samples: usize,
    },

    #[error("voice-activity inference failed")]
    Inference(#[source] Box<dyn std::error::Error + Send + Sync>),

    #[error("voice-activity model returned invalid probability {0}")]
    InvalidProbability(f32),
}