asr-features 0.1.0

ASR feature extraction and cache warming utilities
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

use anyhow::Result;
use sha2::{Digest, Sha256};

use crate::dsp::{compute_audio_featurizer_features, compute_w2v_bert_features};
use crate::util::{hex_prefix, py_bool, py_float};

#[derive(Debug, Clone)]
pub struct AudioFrontendConfig {
    pub sample_rate: u32,
    pub n_fft: usize,
    pub win_length: usize,
    pub hop_length: usize,
    pub n_mels: usize,
    pub preemphasis: f32,
    pub normalize_signal: bool,
    pub normalize_feature: bool,
    pub normalize_per_frame: bool,
}

#[derive(Debug, Clone)]
pub struct W2vBertFrontendConfig {
    pub model_source: String,
    pub sample_rate: u32,
    pub feature_size: usize,
    pub stride: usize,
    pub feature_dim: usize,
    pub padding_value: f32,
}

#[derive(Debug, Clone)]
pub(crate) enum FrontendConfig {
    Audio(AudioFrontendConfig),
    W2vBert(W2vBertFrontendConfig),
}

#[derive(Debug)]
pub struct FeatureMatrix {
    pub rows: usize,
    pub cols: usize,
    pub values: Vec<f32>,
}

pub fn squeezeformer_frontend_config() -> AudioFrontendConfig {
    AudioFrontendConfig {
        sample_rate: 16_000,
        n_fft: 400,
        win_length: 400,
        hop_length: 160,
        n_mels: 80,
        preemphasis: 0.97,
        normalize_signal: true,
        normalize_feature: true,
        normalize_per_frame: false,
    }
}

pub fn zipformer_frontend_config() -> AudioFrontendConfig {
    AudioFrontendConfig {
        sample_rate: 16_000,
        n_fft: 400,
        win_length: 400,
        hop_length: 160,
        n_mels: 80,
        preemphasis: 0.0,
        normalize_signal: false,
        normalize_feature: false,
        normalize_per_frame: false,
    }
}

pub fn w2v_bert_frontend_config(
    model_source: Option<String>,
    sample_rate: Option<u32>,
    feature_size: Option<usize>,
    stride: Option<usize>,
    feature_dim: Option<usize>,
    padding_value: Option<f32>,
) -> W2vBertFrontendConfig {
    let feature_size = feature_size.unwrap_or(80);
    let stride = stride.unwrap_or(2).max(1);
    W2vBertFrontendConfig {
        model_source: model_source.unwrap_or_else(|| "facebook/w2v-bert-2.0".to_string()),
        sample_rate: sample_rate.unwrap_or(16_000),
        feature_size,
        stride,
        feature_dim: feature_dim.unwrap_or(feature_size * stride),
        padding_value: padding_value.unwrap_or(1.0),
    }
}

pub fn extract_audio_features_from_samples(
    waveform: &[f32],
    sample_rate: u32,
    config: &AudioFrontendConfig,
) -> Result<FeatureMatrix> {
    let mut waveform = waveform.to_vec();
    compute_audio_featurizer_features(&mut waveform, sample_rate, config)
}

pub fn extract_w2v_bert_features_from_samples(
    waveform: &[f32],
    sample_rate: u32,
    config: &W2vBertFrontendConfig,
) -> Result<FeatureMatrix> {
    let mut waveform = waveform.to_vec();
    compute_w2v_bert_features(&mut waveform, sample_rate, config)
}

pub(crate) fn compute_features(
    mut waveform: Vec<f32>,
    sample_rate: u32,
    frontend: &FrontendConfig,
) -> Result<FeatureMatrix> {
    match frontend {
        FrontendConfig::Audio(config) => {
            compute_audio_featurizer_features(&mut waveform, sample_rate, config)
        }
        FrontendConfig::W2vBert(config) => {
            compute_w2v_bert_features(&mut waveform, sample_rate, config)
        }
    }
}

impl FrontendConfig {
    pub(crate) fn feature_dim(&self) -> usize {
        match self {
            Self::Audio(config) => config.n_mels,
            Self::W2vBert(config) => config.feature_dim,
        }
    }

    pub(crate) fn sample_rate(&self) -> u32 {
        match self {
            Self::Audio(config) => config.sample_rate,
            Self::W2vBert(config) => config.sample_rate,
        }
    }

    pub(crate) fn config_repr(&self) -> String {
        match self {
            Self::Audio(config) => format!(
                "{{'featurizer': {{'sample_rate': {}, 'n_fft': {}, 'win_length': {}, 'n_mels': {}, 'backend': 'torchaudio', 'preemphasis': {}, 'normalize_signal': {}, 'normalize_feature': {}, 'normalize_per_frame': {}, 'hop_length': {}}}}}",
                config.sample_rate,
                config.n_fft,
                config.win_length,
                config.n_mels,
                py_float(config.preemphasis),
                py_bool(config.normalize_signal),
                py_bool(config.normalize_feature),
                py_bool(config.normalize_per_frame),
                config.hop_length,
            ),
            Self::W2vBert(config) => format!(
                "{{'featurizer': {{'type': 'w2v_bert', 'model_source': '{}', 'sample_rate': {}, 'feature_size': {}, 'stride': {}, 'feature_dim': {}, 'padding_value': {}}}}}",
                config.model_source.replace('\'', "\\'"),
                config.sample_rate,
                config.feature_size,
                config.stride,
                config.feature_dim,
                py_float(config.padding_value),
            ),
        }
    }

    pub(crate) fn frontend_hash(&self) -> String {
        let digest = Sha256::digest(self.config_repr().as_bytes());
        hex_prefix(&digest, 12)
    }
}