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
use std::collections::VecDeque;
use num_complex::Complex64;
#[cfg(feature = "fftrust")]
use num_complex::Complex;
use crate::freqs::{InverseCosineTransform, ForwardRealFourier};
use crate::ringbuffer::Ringbuffer;
#[cfg(feature = "fftextern")]
use fftw::array::AlignedVec;
#[cfg(feature = "fftrust")]
type AlignedVec<T> = Vec<T>;
pub struct Transform {
idct: InverseCosineTransform,
rfft: ForwardRealFourier,
rb: Ringbuffer,
windowed_samples: AlignedVec<f64>,
samples_freq_domain: AlignedVec<Complex64>,
filters: AlignedVec<f64>,
mean_coeffs: Vec<f64>,
prev_coeffs: VecDeque<Vec<f64>>,
maxmel: f64,
sample_rate: usize,
maxfilter: usize,
nfilters: usize,
buffer_size: usize,
normalization_length: usize,
}
impl Transform {
pub fn new(sample_rate: usize, buffer_size: usize) -> Transform {
let size = 2 * buffer_size;
let nfilters = 40;
let maxfilter = 16;
let normalization_length = 5;
#[cfg(feature = "fftrust")]
let windowed_samples = vec![0.0; size];
#[cfg(feature = "fftrust")]
let samples_freq_domain = vec![Complex::i(); size / 2 + 1];
#[cfg(feature = "fftrust")]
let filters = vec![0.0; nfilters];
#[cfg(feature = "fftextern")]
let windowed_samples = AlignedVec::new(size);
#[cfg(feature = "fftextern")]
let samples_freq_domain = AlignedVec::new(size / 2 + 1);
#[cfg(feature = "fftextern")]
let filters = AlignedVec::new(nfilters);
Transform {
idct: InverseCosineTransform::new(nfilters),
rfft: ForwardRealFourier::new(size),
rb: Ringbuffer::new(2 * buffer_size),
windowed_samples,
samples_freq_domain,
filters,
mean_coeffs: vec![0.0; maxfilter*3],
prev_coeffs: VecDeque::new(),
maxmel: 2595.0 * (1.0 + sample_rate as f64 / 2.0 / 700.0).log10(),
sample_rate,
maxfilter,
nfilters,
buffer_size,
normalization_length
}
}
pub fn transform(&mut self, input: &[i16], output: &mut [f64]) {
self.rb.append_back(&input);
self.rb.apply_hamming(&mut self.windowed_samples);
self.rfft.transform(&mut self.windowed_samples, &mut self.samples_freq_domain);
for x in self.filters.iter_mut() {
*x = 0.0;
}
let filter_length = (self.maxmel / self.nfilters as f64) * 2.0;
for (idx, val) in self.samples_freq_domain.iter().skip(1).enumerate() {
let mel = 2595.0 * (1.0 + (self.sample_rate as f64 / 2.0 * (1.0 + idx as f64) / (self.samples_freq_domain.len() as f64)) / 700.0).log10();
let mut idx = ((mel / self.maxmel) * self.nfilters as f64).floor() as usize;
let val = (val.re / self.windowed_samples.len() as f64).powf(2.0) + (val.im / self.windowed_samples.len() as f64).powf(2.0);
if idx == self.nfilters {
idx -= 1;
}
if idx > 0 {
let mel_diff = mel - (idx - 1) as f64 * filter_length / 2.0;
let mel_diff = mel_diff / filter_length;
self.filters[idx-1] += (1.0 - mel_diff) * val;
}
let mel_diff = mel - idx as f64 * filter_length / 2.0;
let mel_diff = mel_diff / filter_length;
self.filters[idx] += mel_diff * val;
}
for filter in self.filters.iter_mut() {
if *filter < 1e-20 {
*filter = -46.05;
} else {
*filter = (*filter).ln();
}
}
self.idct.transform(&mut self.filters, output);
for i in self.maxfilter..output.len() {
output[i] = 0.0;
}
if let Some(back) = self.prev_coeffs.back() {
for i in 0..self.maxfilter {
output[self.maxfilter + i] = output[i] - back[i];
output[self.maxfilter*2 + i] = output[self.maxfilter + i] - back[self.maxfilter + i];
}
}
if self.prev_coeffs.len() < self.normalization_length {
for i in 0..self.maxfilter*3 {
self.mean_coeffs[i] += output[i] / self.normalization_length as f64;
}
} else {
if let Some(front) = self.prev_coeffs.pop_front() {
for i in 0..self.maxfilter*3 {
self.mean_coeffs[i] += (output[i] - front[i]) / self.normalization_length as f64;
}
}
}
self.prev_coeffs.push_back(output.to_vec());
if self.prev_coeffs.len() < self.normalization_length {
return;
}
let mut max_energy = 0.0;
for i in &self.prev_coeffs {
if i[0] > max_energy {
max_energy = i[0];
}
}
for (coeff, mean) in output.iter_mut().zip(self.mean_coeffs.iter()) {
*coeff = *coeff - mean;
}
}
}