ruopus 0.1.2

A pure-Rust implementation of the Opus audio codec (RFC 6716). No FFI; unsafe confined to documented SIMD kernels.
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

//! Decimating resamplers used by the pitch analysis (RFC 6716 §5.2).
//!
//! Pitch estimation runs at 4 kHz (first stage) and 8 kHz (second stage),
//! so the internal-rate frame is decimated by 2 (16→8, 8→4 kHz) with the
//! half-band allpass [`down2`], or by 2/3 (12→8 kHz) with [`down2_3`]. Both
//! are fixed-point and pinned bit-exactly against the compiled reference.

extern crate alloc;
use alloc::vec;

use super::super::math::{rshift_round, smlawb, smulwb};

/// Half-band allpass coefficients for `down2` (Q15-ish).
const DOWN2_0: i32 = 9872;
const DOWN2_1: i32 = 39809 - 65536; // -25727

/// 2/3 resampler coefficients: AR coefficients (`[0..2]`, Q14) and FIR
/// interpolation coefficients (`[2..6]`).
const COEFS_2_3: [i32; 6] = [-2797, -6507, 4697, 10739, 1567, 8276];
const ORDER_FIR: usize = 4;
/// Maximum input batch size (10 ms × 48 kHz).
const MAX_BATCH_SIZE_IN: usize = 480;

/// Saturate to i16.
fn sat16(a: i32) -> i16 {
    a.clamp(i32::from(i16::MIN), i32::from(i16::MAX)) as i16
}

/// Decimate `inp` by two through a second-order
/// allpass half-band filter. `out` receives `inp.len()/2` samples; `s` is
/// the 2-element state carried across calls.
pub(crate) fn down2(s: &mut [i32; 2], out: &mut [i16], inp: &[i16]) {
    let len2 = inp.len() / 2;
    for k in 0..len2 {
        let in32 = i32::from(inp[2 * k]) << 10;
        let y = in32.wrapping_sub(s[0]);
        let x = smlawb(y, y, DOWN2_1);
        let mut out32 = s[0].wrapping_add(x);
        s[0] = in32.wrapping_add(x);

        let in32 = i32::from(inp[2 * k + 1]) << 10;
        let y = in32.wrapping_sub(s[1]);
        let x = smulwb(y, DOWN2_0);
        out32 = out32.wrapping_add(s[1]).wrapping_add(x);
        s[1] = in32.wrapping_add(x);

        out[k] = sat16(rshift_round(out32, 11));
    }
}

/// Second-order AR filter, output in Q8.
fn ar2(s: &mut [i32], out_q8: &mut [i32], inp: &[i16], a_q14: &[i32]) {
    for (k, &xn) in inp.iter().enumerate() {
        let out32 = s[0].wrapping_add(i32::from(xn) << 8);
        out_q8[k] = out32;
        let out32 = out32 << 2;
        s[0] = smlawb(s[1], out32, a_q14[0]);
        s[1] = smulwb(out32, a_q14[1]);
    }
}

/// Decimate `inp` by 2/3 (12→8 kHz). `out`
/// receives `floor(2*inp.len()/3)` samples; `s` is the 6-element state.
pub(crate) fn down2_3(s: &mut [i32; 6], out: &mut [i16], inp: &[i16]) {
    let mut buf = vec![0i32; MAX_BATCH_SIZE_IN + ORDER_FIR];
    buf[..ORDER_FIR].copy_from_slice(&s[..ORDER_FIR]);

    let mut in_off = 0usize;
    let mut in_len = inp.len();
    let mut out_off = 0usize;
    let mut n_samples;

    loop {
        n_samples = in_len.min(MAX_BATCH_SIZE_IN);
        // AR2 state lives in s[ORDER_FIR..ORDER_FIR+2]; the FIR history in
        // s[0..ORDER_FIR] is untouched here. Output into buf[ORDER_FIR..].
        ar2(
            &mut s[ORDER_FIR..ORDER_FIR + 2],
            &mut buf[ORDER_FIR..ORDER_FIR + n_samples],
            &inp[in_off..in_off + n_samples],
            &COEFS_2_3,
        );

        let mut bp = 0usize;
        let mut counter = n_samples as i32;
        while counter > 2 {
            let mut r = smulwb(buf[bp], COEFS_2_3[2]);
            r = smlawb(r, buf[bp + 1], COEFS_2_3[3]);
            r = smlawb(r, buf[bp + 2], COEFS_2_3[5]);
            r = smlawb(r, buf[bp + 3], COEFS_2_3[4]);
            out[out_off] = sat16(rshift_round(r, 6));
            out_off += 1;

            let mut r = smulwb(buf[bp + 1], COEFS_2_3[4]);
            r = smlawb(r, buf[bp + 2], COEFS_2_3[5]);
            r = smlawb(r, buf[bp + 3], COEFS_2_3[3]);
            r = smlawb(r, buf[bp + 4], COEFS_2_3[2]);
            out[out_off] = sat16(rshift_round(r, 6));
            out_off += 1;

            bp += 3;
            counter -= 3;
        }

        in_off += n_samples;
        in_len -= n_samples;
        if in_len > 0 {
            buf.copy_within(n_samples..n_samples + ORDER_FIR, 0);
        } else {
            break;
        }
    }
    s[..ORDER_FIR].copy_from_slice(&buf[n_samples..n_samples + ORDER_FIR]);
}

#[cfg(test)]
mod tests {
    use alloc::vec;

    use super::*;

    fn pin_input() -> [i16; 48] {
        core::array::from_fn(|i| ((i as i32 * 1237 + 11) % 9001 - 4500) as i16)
    }

    /// Bit-exact pin against the reference.
    #[test]
    fn down2_matches_reference_pin() {
        let inp = pin_input();
        let mut s = [0i32; 2];
        let mut out = [0i16; 24];
        down2(&mut s, &mut out, &inp);
        let expected: [i16; 24] = [
            -1608, -3676, 351, 2063, 1561, -3464, 1555, 2108, -1840, -1968, 2388, 262, -3774, 644, 2240, 1790, -3263,
            1773, 2315, -1627, -1758, 2600, 472, -3562,
        ];
        assert_eq!(out, expected, "down2 output disagrees with reference");
        assert_eq!(s, [1_850_487, 176_361], "down2 state disagrees with reference");
    }

    /// Bit-exact pin against the reference.
    #[test]
    fn down2_3_matches_reference_pin() {
        let inp = pin_input();
        let mut s = [0i32; 6];
        let mut out = [0i16; 32];
        down2_3(&mut s, &mut out, &inp);
        let expected: [i16; 32] = [
            0, -1287, -4180, -1508, 499, 1945, 3188, -2326, -2463, 1251, 2064, 1704, -4002, -1061, 1242, 2366, -605,
            -4390, -8, 980, 2852, 2250, -3529, -466, 1757, 2921, -62, -3847, 538, 1524, 2536, -2665,
        ];
        assert_eq!(out, expected, "down2_3 output disagrees with reference");
    }

    /// A low-frequency tone passes through `down2`; a near-Nyquist tone is
    /// attenuated (the half-band filter's purpose).
    #[test]
    fn down2_attenuates_high_frequencies() {
        let n = 480;
        let low: alloc::vec::Vec<i16> = (0..n).map(|i| ((i as f32 * 0.05).sin() * 10000.0) as i16).collect();
        let high: alloc::vec::Vec<i16> = (0..n)
            .map(|i| ((i as f32 * core::f32::consts::PI).sin() * 10000.0) as i16)
            .collect();
        let mut s = [0i32; 2];
        let mut out_low = vec![0i16; n / 2];
        down2(&mut s, &mut out_low, &low);
        let mut s = [0i32; 2];
        let mut out_high = vec![0i16; n / 2];
        down2(&mut s, &mut out_high, &high);
        let e_low: i64 = out_low.iter().map(|&v| i64::from(v) * i64::from(v)).sum();
        let e_high: i64 = out_high.iter().map(|&v| i64::from(v) * i64::from(v)).sum();
        assert!(
            e_low > 10 * e_high,
            "high freq not attenuated: low {e_low} high {e_high}"
        );
    }
}