audiofp 0.3.8

Pure-Rust audio fingerprinting and identification: Wang, Panako, Haitsma–Kalker, ONNX neural embedder, AudioSeal watermark, and streaming variants. no_std + alloc capable, bytemuck-friendly hash types.
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

//! Extended codec and resampler robustness tests.
//!
//! Adds to the base `codec_roundtrip.rs`:
//! - AIFF format decoding
//! - Two-track identification (Galway vs Furious Freak)
//! - Mono vs stereo downmix equivalence
//! - MP3 sample-rate ladder (8kHz–44.1kHz → 8kHz fingerprint)
//!
//! Audio: Kevin MacLeod (incompetech.com), CC-BY 3.0. See CREDITS.md.
#![cfg(feature = "std")]

use std::collections::HashSet;

use audiofp::classical::{Panako, Wang};
use audiofp::io::decode_to_mono_at;
use audiofp::{AudioBuffer, Fingerprinter, SampleRate};

fn wang_hashes(path: &str) -> HashSet<u32> {
    let samples =
        decode_to_mono_at(path, 8_000).unwrap_or_else(|e| panic!("failed to decode {path}: {e}"));
    let mut wang = Wang::default();
    wang.extract(AudioBuffer::new(&samples, SampleRate::HZ_8000))
        .unwrap()
        .hashes
        .into_iter()
        .map(|h| h.hash)
        .collect()
}

fn jaccard(a: &HashSet<u32>, b: &HashSet<u32>) -> f32 {
    let union = a.union(b).count();
    if union == 0 {
        return 0.0;
    }
    a.intersection(b).count() as f32 / union as f32
}

// ═══════════════════════════════════════════════════════════════════════════
// AIFF format support
// ═══════════════════════════════════════════════════════════════════════════

#[test]
fn aiff_decodes_and_matches_flac() {
    let ref_h = wang_hashes("tests/assets/galway.flac");
    let aiff_h = wang_hashes("tests/assets/galway.aiff");
    let j = jaccard(&ref_h, &aiff_h);
    eprintln!("Wang AIFF vs FLAC: Jaccard = {j:.3}");
    // AIFF is lossless PCM in a different container — should be ~identical
    assert!(j >= 0.95, "AIFF vs FLAC Jaccard = {j:.3} (expected ≥0.95)");
}

// ═══════════════════════════════════════════════════════════════════════════
// Two-track identification: Galway vs Furious Freak
// ═══════════════════════════════════════════════════════════════════════════

#[test]
fn same_track_different_codecs_high_overlap() {
    // Furious Freak in MP3 vs FLAC — same song, different codec
    let freak_flac = wang_hashes("tests/assets/freak.flac");
    let freak_mp3 = wang_hashes("tests/assets/freak.mp3");
    let j = jaccard(&freak_flac, &freak_mp3);
    eprintln!("Freak MP3 vs FLAC: Jaccard = {j:.3}");
    assert!(
        j >= 0.20,
        "Same track cross-codec Jaccard = {j:.3} (expected ≥0.20)"
    );
}

#[test]
fn different_tracks_near_zero_overlap() {
    // Galway vs Furious Freak — different songs, should have near-zero overlap
    let galway = wang_hashes("tests/assets/galway.flac");
    let freak = wang_hashes("tests/assets/freak.flac");
    let j = jaccard(&galway, &freak);
    eprintln!("Galway vs Freak: Jaccard = {j:.3} (should be near 0)");
    // Different songs should have <5% hash overlap (random collision level)
    assert!(
        j < 0.05,
        "Different tracks overlap = {j:.3} (expected <0.05 — fingerprints not discriminating!)"
    );
}

#[test]
fn identification_scenario_cross_codec() {
    // The real scenario: can we identify Galway-as-MP3 against a database
    // containing Galway-as-FLAC and Freak-as-FLAC?
    let query = wang_hashes("tests/assets/galway.mp3");
    let db_galway = wang_hashes("tests/assets/galway.flac");
    let db_freak = wang_hashes("tests/assets/freak.flac");

    let overlap_galway = jaccard(&query, &db_galway);
    let overlap_freak = jaccard(&query, &db_freak);

    eprintln!("Query=galway.mp3 vs DB galway.flac: {overlap_galway:.3}");
    eprintln!("Query=galway.mp3 vs DB freak.flac:  {overlap_freak:.3}");

    // The correct match should be significantly higher
    assert!(
        overlap_galway > overlap_freak * 5.0,
        "Identification failed: galway overlap ({overlap_galway:.3}) not >> freak overlap ({overlap_freak:.3})"
    );
}

// ═══════════════════════════════════════════════════════════════════════════
// Mono vs stereo downmix equivalence
// ═══════════════════════════════════════════════════════════════════════════

#[test]
fn mono_vs_stereo_same_track_high_overlap() {
    let mono_h = wang_hashes("tests/assets/galway.mp3");
    let stereo_h = wang_hashes("tests/assets/galway_stereo.mp3");
    let j = jaccard(&mono_h, &stereo_h);
    eprintln!("Wang mono vs stereo MP3: Jaccard = {j:.3}");
    // Stereo downmix of the same track should preserve most peaks.
    // Note: ESP-ADF stereo variant uses joint-stereo encoding which
    // allocates bits differently than the mono encode, so overlap is
    // lower than lossless container differences.
    assert!(
        j >= 0.25,
        "Mono vs stereo Jaccard = {j:.3} (expected ≥0.25)"
    );
}

#[test]
fn mono_vs_stereo_flac() {
    let mono_h = wang_hashes("tests/assets/galway.flac");
    let stereo_h = wang_hashes("tests/assets/galway_stereo.flac");
    let j = jaccard(&mono_h, &stereo_h);
    eprintln!("Wang mono vs stereo FLAC: Jaccard = {j:.3}");
    assert!(
        j >= 0.60,
        "Mono vs stereo FLAC Jaccard = {j:.3} (expected ≥0.60)"
    );
}

// ═══════════════════════════════════════════════════════════════════════════
// MP3 sample-rate ladder: fingerprint survives arbitrary source rates
// ═══════════════════════════════════════════════════════════════════════════

#[test]
fn sample_rate_ladder_all_produce_hashes() {
    let rates = [
        ("tests/assets/freak_8000hz.mp3", 8000),
        ("tests/assets/freak_11025hz.mp3", 11025),
        ("tests/assets/freak_16000hz.mp3", 16000),
        ("tests/assets/freak_22050hz.mp3", 22050),
        ("tests/assets/freak_32000hz.mp3", 32000),
        ("tests/assets/freak_44100hz.mp3", 44100),
    ];

    for (path, native_sr) in rates {
        let samples = decode_to_mono_at(path, 8_000)
            .unwrap_or_else(|e| panic!("{path} (native {native_sr} Hz) decode failed: {e}"));
        let mut wang = Wang::default();
        let fp = wang
            .extract(AudioBuffer::new(&samples, SampleRate::HZ_8000))
            .unwrap_or_else(|e| panic!("{path} extract failed: {e}"));
        assert!(
            fp.hashes.len() > 20,
            "{path} (native {native_sr} Hz): only {} hashes (expected >20)",
            fp.hashes.len()
        );
        eprintln!("{path} ({native_sr} Hz → 8kHz): {} hashes", fp.hashes.len());
    }
}

#[test]
fn sample_rate_ladder_overlap_with_reference() {
    // All sample-rate variants are the same song (Furious Freak).
    // When resampled to 8kHz and fingerprinted, they should have
    // meaningful overlap with the 44.1kHz reference.
    let ref_h = wang_hashes("tests/assets/freak_44100hz.mp3");

    let variants = [
        "tests/assets/freak_8000hz.mp3",
        "tests/assets/freak_11025hz.mp3",
        "tests/assets/freak_16000hz.mp3",
        "tests/assets/freak_22050hz.mp3",
        "tests/assets/freak_32000hz.mp3",
    ];

    for path in variants {
        let h = wang_hashes(path);
        let j = jaccard(&ref_h, &h);
        eprintln!("{path} vs 44.1kHz ref: Jaccard = {j:.3}");
        // Low source rates (8kHz) lose high-frequency peaks, so lower threshold
        assert!(
            j >= 0.05,
            "{path}: Jaccard vs 44.1kHz = {j:.3} (expected ≥0.05)"
        );
    }
}

#[test]
fn panako_sample_rate_ladder() {
    // Panako should also handle the sample-rate ladder
    let ref_samples = decode_to_mono_at("tests/assets/freak_44100hz.mp3", 8_000).unwrap();
    let mut panako = Panako::default();
    let ref_fp = panako
        .extract(AudioBuffer::new(&ref_samples, SampleRate::HZ_8000))
        .unwrap();
    let ref_set: HashSet<u32> = ref_fp.hashes.iter().map(|h| h.hash).collect();

    let variants = [
        "tests/assets/freak_8000hz.mp3",
        "tests/assets/freak_22050hz.mp3",
        "tests/assets/freak_32000hz.mp3",
    ];

    for path in variants {
        let samples = decode_to_mono_at(path, 8_000).unwrap();
        let fp = panako
            .extract(AudioBuffer::new(&samples, SampleRate::HZ_8000))
            .unwrap();
        let h: HashSet<u32> = fp.hashes.iter().map(|h| h.hash).collect();
        let j = jaccard(&ref_set, &h);
        eprintln!("Panako {path} vs 44.1kHz: Jaccard = {j:.3}");
        assert!(
            fp.hashes.len() > 20,
            "Panako {path}: only {} hashes",
            fp.hashes.len()
        );
    }
}