voxudio 0.7.1

A real-time audio processing library with ONNX runtime support
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

use {std::f32::consts::PI, voxudio::*};

#[tokio::test]
async fn test_tcc_creation() {
    let tcc = ToneColorConverter::new("../checkpoint/tone_color_converter.onnx");
    assert!(tcc.is_ok(), "Should be able to create ToneColorConverter");
}

#[tokio::test]
async fn test_tcc_process() -> anyhow::Result<()> {
    let mut tcc = ToneColorConverter::new("../checkpoint/tone_color_converter.onnx")?;

    // Create a simulated speech signal
    let sample_rate = 16000; // 16kHz
    let duration_secs = 3.0;
    let num_samples = (sample_rate as f32 * duration_secs) as usize;

    let mut source_audio = Vec::with_capacity(num_samples);

    // Create a signal with multiple frequencies to simulate speech
    for i in 0..num_samples {
        let t = i as f32 / sample_rate as f32;
        // Fundamental frequency + harmonics to simulate speech characteristics
        let f0 = 120.0; // Fundamental frequency, similar to human voice
        let sample = 0.5 * (2.0 * PI * f0 * t).sin()
            + 0.3 * (2.0 * PI * f0 * 2.0 * t).sin()
            + 0.2 * (2.0 * PI * f0 * 3.0 * t).sin();

        // Add amplitude variation to simulate speech prosody
        let envelope = 0.5 + 0.5 * (2.0 * PI * 0.5 * t).sin();

        source_audio.push(sample * envelope * 0.5);
    }

    // Create source and target speaker feature vectors (256-dim)
    let source_features = [[0.0f32; 256]];
    let target_features = [[0.0f32; 256]];

    // Process audio data
    let converted_audio = tcc
        .convert(&source_audio, &source_features, &target_features)
        .await?;

    // Verify converted audio
    // Converted audio length should be similar to source
    assert!(
        (converted_audio.0.len() as i32 - source_audio.len() as i32).abs() < 1000,
        "Converted audio length should be close to source length"
    );

    // Converted audio shouldn't be all zeros
    let sum: f32 = converted_audio.0.iter().map(|&x| x.abs()).sum();
    assert!(sum > 0.0, "Converted audio shouldn't be all zeros");

    Ok(())
}

#[tokio::test]
async fn test_tcc_with_real_features() -> anyhow::Result<()> {
    // This test uses real feature extractor to generate target features
    let mut tcc = ToneColorConverter::new("../checkpoint/tone_color_converter.onnx")?;
    let mut see = SpeakerEmbeddingExtractor::new("../checkpoint/speaker_embedding_extractor.onnx")?;

    // Create two different simulated speech signals
    let sample_rate = 16000;
    let duration_secs = 3.0;
    let num_samples = (sample_rate as f32 * duration_secs) as usize;

    // Source audio - lower fundamental frequency (male voice)
    let mut source_audio = Vec::with_capacity(num_samples);
    for i in 0..num_samples {
        let t = i as f32 / sample_rate as f32;
        let f0 = 120.0; // Lower fundamental frequency
        let sample = 0.5 * (2.0 * PI * f0 * t).sin()
            + 0.3 * (2.0 * PI * f0 * 2.0 * t).sin()
            + 0.2 * (2.0 * PI * f0 * 3.0 * t).sin();
        let envelope = 0.5 + 0.5 * (2.0 * PI * 0.5 * t).sin();
        source_audio.push(sample * envelope * 0.5);
    }

    // Target audio - higher fundamental frequency (female voice)
    let mut target_audio = Vec::with_capacity(num_samples);
    for i in 0..num_samples {
        let t = i as f32 / sample_rate as f32;
        let f0 = 220.0; // 较高的基频
        let sample = 0.5 * (2.0 * PI * f0 * t).sin()
            + 0.3 * (2.0 * PI * f0 * 2.0 * t).sin()
            + 0.2 * (2.0 * PI * f0 * 3.0 * t).sin();
        let envelope = 0.5 + 0.5 * (2.0 * PI * 0.5 * t).sin();
        target_audio.push(sample * envelope * 0.5);
    }

    // Extract features from source and target audio
    let source_features = see.extract(&source_audio, 1).await?;
    let target_features = see.extract(&target_audio, 1).await?;

    // Perform voice conversion using extracted features
    let converted_audio = tcc
        .convert(&source_audio, &source_features, &target_features)
        .await?;

    // 转换后的音频长度应该与源音频相同或接近
    assert!(
        (converted_audio.0.len() as i32 - source_audio.len() as i32).abs() < 1000,
        "Converted audio length should be close to source length"
    );

    // 转换后的音频不应该全是零
    let sum: f32 = converted_audio.0.iter().map(|&x| x.abs()).sum();
    assert!(sum > 0.0, "Converted audio shouldn't be all zeros");

    Ok(())
}

#[tokio::test]
async fn test_tcc_empty_input() -> anyhow::Result<()> {
    let mut tcc = ToneColorConverter::new("../checkpoint/tone_color_converter.onnx")?;

    // Test empty input
    let empty_audio = Vec::<f32>::new();
    let source_features = vec![0.0f32; 256];
    let target_features = vec![0.0f32; 256];

    let source_features_array: Vec<[f32; 256]> = source_features
        .chunks(256)
        .map(|chunk| {
            let mut arr = [0.0; 256];
            arr.copy_from_slice(chunk);
            arr
        })
        .collect();

    let target_features_array: Vec<[f32; 256]> = target_features
        .chunks(256)
        .map(|chunk| {
            let mut arr = [0.0; 256];
            arr.copy_from_slice(chunk);
            arr
        })
        .collect();

    let result = tcc
        .convert(&empty_audio, &source_features_array, &target_features_array)
        .await;

    // 空输入应该返回错误
    assert!(
        result.is_err(),
        "Processing empty input should return error"
    );

    Ok(())
}

#[tokio::test]
async fn test_tcc_feature_influence() -> anyhow::Result<()> {
    let mut tcc = ToneColorConverter::new("../checkpoint/tone_color_converter.onnx")?;

    // Create a simple audio signal
    let sample_rate = 16000;
    let duration_secs = 3.0;
    let num_samples = (sample_rate as f32 * duration_secs) as usize;

    let mut audio_data = Vec::with_capacity(num_samples);
    for i in 0..num_samples {
        let t = i as f32 / sample_rate as f32;
        let f0 = 150.0;
        let sample = 0.5 * (2.0 * PI * f0 * t).sin();
        audio_data.push(sample);
    }

    // Create two different feature vectors
    let mut features1 = [[0.0f32; 256]];
    let mut features2 = [[0.0f32; 256]];

    // Feature vector 1: all set to 0.1
    for i in 0..256 {
        features1[0][i] = 0.1;
    }

    // Feature vector 2: all set to -0.1
    for i in 0..256 {
        features2[0][i] = -0.1;
    }

    // 使用两个不同的特征向量进行转换
    let converted1 = tcc.convert(&audio_data, &features1, &features1).await?;
    let converted2 = tcc.convert(&audio_data, &features2, &features2).await?;

    // 验证两个不同特征向量产生的结果应该不同
    // 计算两个转换结果的差异
    let mut diff_sum = 0.0;
    let min_len = converted1.0.len().min(converted2.0.len());

    for i in 0..min_len {
        diff_sum += (converted1.0[i] - converted2.0[i]).abs();
    }

    // 不同的特征向量应该产生不同的转换结果
    assert!(
        diff_sum > 0.0,
        "Different feature vectors should produce different conversion results"
    );

    Ok(())
}