mousiki 0.2.0

Pure Rust Opus codec.
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

//! Port of `silk_process_gains_FLP`.
//!
//! Mirrors the floating-point gain processing helper from
//! `silk/float/process_gains_FLP.c`, applying the LTP-dependent gain reduction,
//! limiting the subframe gains, quantising them, and updating the lambda and
//! quantisation-offset bookkeeping for the FLP encoder path.

use crate::silk::decode_indices::ConditionalCoding;
use crate::silk::encoder::control_flp::EncoderControlFlp;
use crate::silk::encoder::state_flp::EncoderStateFlp;
use crate::silk::gain_quant::silk_gains_quant;
use crate::silk::sigproc_flp::{silk_min_float, silk_sigmoid};
use crate::silk::tables_other::SILK_QUANTIZATION_OFFSETS_Q10;
use crate::silk::tuning_parameters::{
    LAMBDA_CODING_QUALITY, LAMBDA_DELAYED_DECISIONS, LAMBDA_INPUT_QUALITY, LAMBDA_OFFSET,
    LAMBDA_QUANT_OFFSET, LAMBDA_SPEECH_ACT,
};
use crate::silk::{FrameQuantizationOffsetType, FrameSignalType, MAX_NB_SUBFR};
use libm::{powf, sqrtf};

const Q16_SCALE: f32 = 65_536.0;

/// Mirrors `silk_process_gains_FLP`.
pub fn process_gains_flp(
    encoder: &mut EncoderStateFlp,
    control: &mut EncoderControlFlp,
    cond_coding: ConditionalCoding,
) {
    let nb_subfr = encoder.common.nb_subfr;
    assert!(
        nb_subfr == MAX_NB_SUBFR || nb_subfr == MAX_NB_SUBFR / 2,
        "encoder supports 2 or 4 subframes"
    );

    if matches!(encoder.common.indices.signal_type, FrameSignalType::Voiced) {
        let reduction =
            1.0_f32 - 0.5_f32 * silk_sigmoid(0.25_f32 * (control.lt_pred_cod_gain - 12.0_f32));
        for gain in control.gains.iter_mut().take(nb_subfr) {
            *gain *= reduction;
        }
    }

    let subfr_length = encoder.common.subfr_length as f32;
    assert!(subfr_length > 0.0, "subframe length must be positive");

    let inv_max_sqr_val = powf(
        2.0,
        0.33 * (21.0 - (encoder.common.snr_db_q7 as f32) * (1.0 / 128.0)),
    ) / subfr_length;

    let res_nrg = control.res_nrg;
    for (k, gain) in control.gains.iter_mut().take(nb_subfr).enumerate() {
        let adjusted = sqrtf(*gain * *gain + res_nrg[k] * inv_max_sqr_val);
        *gain = silk_min_float(adjusted, 32_767.0);
    }

    let mut gains_q16 = [0i32; MAX_NB_SUBFR];
    for (dst, &gain) in gains_q16
        .iter_mut()
        .zip(control.gains.iter().take(nb_subfr))
    {
        *dst = (gain * Q16_SCALE) as i32;
    }

    control.gains_unq_q16[..nb_subfr].copy_from_slice(&gains_q16[..nb_subfr]);
    control.last_gain_index_prev = encoder.shape_state.last_gain_index;

    let conditional = matches!(cond_coding, ConditionalCoding::Conditional);
    let mut last_gain_index = control.last_gain_index_prev;
    silk_gains_quant(
        &mut encoder.common.indices.gains_indices[..nb_subfr],
        &mut gains_q16[..nb_subfr],
        &mut last_gain_index,
        conditional,
    );
    encoder.shape_state.last_gain_index = last_gain_index;

    for (gain, &quant_q16) in control
        .gains
        .iter_mut()
        .zip(gains_q16.iter())
        .take(nb_subfr)
    {
        *gain = quant_q16 as f32 / Q16_SCALE;
    }

    if matches!(encoder.common.indices.signal_type, FrameSignalType::Voiced) {
        let combined =
            control.lt_pred_cod_gain + (encoder.common.input_tilt_q15 as f32) * (1.0 / 32_768.0);
        encoder.common.indices.quant_offset_type = if combined > 1.0 {
            FrameQuantizationOffsetType::Low
        } else {
            FrameQuantizationOffsetType::High
        };
    }

    let signal_row = (i32::from(encoder.common.indices.signal_type) >> 1) as usize;
    let quant_col = match encoder.common.indices.quant_offset_type {
        FrameQuantizationOffsetType::Low => 0,
        FrameQuantizationOffsetType::High => 1,
    };
    let quant_offset =
        f32::from(SILK_QUANTIZATION_OFFSETS_Q10[signal_row][quant_col]) * (1.0 / 1024.0);

    control.lambda = LAMBDA_OFFSET
        + LAMBDA_DELAYED_DECISIONS * encoder.common.n_states_delayed_decision as f32
        + LAMBDA_SPEECH_ACT * encoder.common.speech_activity_q8 as f32 * (1.0 / 256.0)
        + LAMBDA_INPUT_QUALITY * control.input_quality
        + LAMBDA_CODING_QUALITY * control.coding_quality
        + LAMBDA_QUANT_OFFSET * quant_offset;

    debug_assert!(
        (0.0..2.0).contains(&control.lambda),
        "lambda must remain inside the reference range"
    );
}

#[cfg(test)]
mod tests {
    use super::*;

    #[test]
    fn process_gains_flp_updates_indices_and_lambda() {
        let mut encoder = EncoderStateFlp::default();
        encoder.common.nb_subfr = 2;
        encoder.common.indices.signal_type = FrameSignalType::Voiced;
        encoder.common.n_states_delayed_decision = 1;
        encoder.common.speech_activity_q8 = 128;
        encoder.common.input_tilt_q15 = 5_000;

        let mut control = EncoderControlFlp::default();
        control.gains[0] = 2.0;
        control.gains[1] = 3.0;
        control.lt_pred_cod_gain = 0.5;
        control.res_nrg[0] = 0.0;
        control.res_nrg[1] = 0.0;
        control.input_quality = 0.4;
        control.coding_quality = 0.2;

        let nb_subfr = encoder.common.nb_subfr;
        let reduction = 1.0 - 0.5 * silk_sigmoid(0.25 * (control.lt_pred_cod_gain - 12.0));
        let mut expected_gains = control.gains;
        for gain in expected_gains.iter_mut().take(nb_subfr) {
            *gain *= reduction;
        }

        let inv_max_sqr_val = powf(
            2.0,
            0.33 * (21.0 - (encoder.common.snr_db_q7 as f32) * (1.0 / 128.0)),
        ) / encoder.common.subfr_length as f32;
        for (k, gain) in expected_gains.iter_mut().take(nb_subfr).enumerate() {
            let clamped = sqrtf(*gain * *gain + control.res_nrg[k] * inv_max_sqr_val);
            *gain = silk_min_float(clamped, 32_767.0);
        }

        let mut expected_unquant_q16 = [0i32; MAX_NB_SUBFR];
        for (dst, &gain) in expected_unquant_q16
            .iter_mut()
            .zip(expected_gains.iter().take(nb_subfr))
        {
            *dst = (gain * Q16_SCALE) as i32;
        }
        let mut expected_quant_q16 = expected_unquant_q16;
        let mut expected_indices = encoder.common.indices.gains_indices;
        let mut expected_last = encoder.shape_state.last_gain_index;
        silk_gains_quant(
            &mut expected_indices[..nb_subfr],
            &mut expected_quant_q16[..nb_subfr],
            &mut expected_last,
            false,
        );

        let expected_quant_offset = {
            let combined = control.lt_pred_cod_gain
                + (encoder.common.input_tilt_q15 as f32) * (1.0 / 32_768.0);
            let quant_offset_type = if combined > 1.0 {
                FrameQuantizationOffsetType::Low
            } else {
                FrameQuantizationOffsetType::High
            };
            let quant_col = match quant_offset_type {
                FrameQuantizationOffsetType::Low => 0,
                FrameQuantizationOffsetType::High => 1,
            };
            let row = (i32::from(encoder.common.indices.signal_type) >> 1) as usize;
            f32::from(SILK_QUANTIZATION_OFFSETS_Q10[row][quant_col]) * (1.0 / 1024.0)
        };
        let expected_lambda = LAMBDA_OFFSET
            + LAMBDA_DELAYED_DECISIONS * encoder.common.n_states_delayed_decision as f32
            + LAMBDA_SPEECH_ACT * encoder.common.speech_activity_q8 as f32 * (1.0 / 256.0)
            + LAMBDA_INPUT_QUALITY * control.input_quality
            + LAMBDA_CODING_QUALITY * control.coding_quality
            + LAMBDA_QUANT_OFFSET * expected_quant_offset;

        process_gains_flp(&mut encoder, &mut control, ConditionalCoding::Independent);

        assert_eq!(
            control.gains_unq_q16[..nb_subfr],
            expected_unquant_q16[..nb_subfr]
        );
        assert_eq!(
            encoder.common.indices.gains_indices[..nb_subfr],
            expected_indices[..nb_subfr]
        );
        assert_eq!(encoder.shape_state.last_gain_index, expected_last);
        for (actual, expected) in control
            .gains
            .iter()
            .take(nb_subfr)
            .zip(expected_quant_q16.iter())
        {
            assert!((actual - (*expected as f32) / Q16_SCALE).abs() < 1e-6);
        }
        assert_eq!(
            encoder.common.indices.quant_offset_type,
            FrameQuantizationOffsetType::High
        );
        assert!((control.lambda - expected_lambda).abs() < 1e-6);
    }
}