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
225
226

pub trait Analyze {
    fn yin_pitch_detection(&mut self, sample_rate: f32, threshold: f32) -> (f32, f32);
    fn get_better_tau(&mut self, tau: usize, sample_rate: f32) -> f32;
    fn yin_difference(&mut self);
    fn yin_absolute_threshold(&mut self, threshold: f32) -> Option<usize>;
    fn yin_parabolic_interpolation(&mut self, tau_estimate: usize) -> f32;
    fn yin_cumulative_mean_normalized_difference(&mut self);
    fn gain(&mut self) -> f32;
    fn analyze(&mut self, sample_rate: f32, threshold: f32) -> DetectionResult;
}

pub struct DetectionResult {
    pub frequency: f32,
    pub probability: f32,
    pub gain: f32,
}

impl Analyze for Vec<f32> {
    fn analyze(&mut self, sample_rate: f32, threshold: f32) -> DetectionResult {
        let gain = self.gain();
        let (frequency, probability) = self.yin_pitch_detection(sample_rate, threshold);

        DetectionResult {
            frequency,
            probability,
            gain,
        }
    }

    fn gain(&mut self) -> f32 {
        let mean_squared: f32 = self.iter().cloned().fold(0.0, |mut sum, x: f32| {
            sum += x.powi(2);
            sum
        });

        let root_mean_squared = mean_squared.sqrt() / 100.0;
        if root_mean_squared < 1.0 {
            root_mean_squared
        } else {
            1.0
        }
    }

    fn yin_pitch_detection(&mut self, sample_rate: f32, threshold: f32) -> (f32, f32) {
        for sample in self.iter_mut() {
            *sample *= 1000.0;
        }

        self.yin_difference();
        self.yin_cumulative_mean_normalized_difference();
        let (probability, pitch_in_hertz) =
            if let Some(tau) = self.yin_absolute_threshold(threshold) {
                (1.0 - self[tau], self.get_better_tau(tau, sample_rate))
            } else {
                (-1.0, 0.0)
            };

        (pitch_in_hertz, probability)
    }

    fn get_better_tau(&mut self, tau: usize, sample_rate: f32) -> f32 {
        let better_tau = self.yin_parabolic_interpolation(tau);
        let pitch_in_hertz = sample_rate / better_tau;
        pitch_in_hertz
    }

    fn yin_difference(&mut self) {
        let buffer_clone = self.clone();
        let half_buffer_size = self.len() / 2;

        for tau in 0..half_buffer_size {
            for i in 0..half_buffer_size {
                let delta: f32 = buffer_clone[i] - buffer_clone[i + tau];
                self[tau] += delta * delta;
            }
        }

        self.resize(half_buffer_size, 0.0)
    }

    fn yin_cumulative_mean_normalized_difference(&mut self) {
        let buffer_size = self.len();
        let mut running_sum: f32 = 0.0;

        for tau in 1..buffer_size {
            running_sum += self[tau];
            self[tau] *= tau as f32 / running_sum;
        }
    }

    fn yin_absolute_threshold(&mut self, threshold: f32) -> Option<usize> {
        let mut iter = self.iter()
            .enumerate()
            .skip(2)
            .skip_while(|(_, &sample)| sample > threshold);
        let tripped_threshold = iter.next()?;

        let (_, mut previous_sample) = tripped_threshold;
        for (index, sample) in iter {
            if sample > previous_sample {
                return Some(index - 1);
            };
            previous_sample = sample;
        }

        Some(self.len() - 1)
    }

    fn yin_parabolic_interpolation(&mut self, tau_estimate: usize) -> f32 {
        let better_tau: f32;

        let x0: usize = if tau_estimate < 1 {
            tau_estimate
        } else {
            tau_estimate - 1
        };

        let x2: usize = if tau_estimate + 1 < self.len() {
            tau_estimate + 1
        } else {
            tau_estimate
        };

        if x0 == tau_estimate {
            better_tau = if self[tau_estimate] <= self[x2] {
                tau_estimate as f32
            } else {
                x2 as f32
            }
        } else if x2 == tau_estimate {
            better_tau = if self[tau_estimate] <= self[x0] {
                tau_estimate as f32
            } else {
                x0 as f32
            }
        } else {
            let s0: f32 = self[x0];
            let s1: f32 = self[tau_estimate];
            let s2: f32 = self[x2];

            better_tau = tau_estimate as f32 + (s2 - s0) / (2.0 * (2.0 * s1 - s2 - s0));
        }

        better_tau
    }
}

#[cfg(test)]
mod tests {
    use analyze::*;
    #[test]
    fn gain_test() {
        let mut buffer = vec![
            0.0, 0.06279052, 0.12533323, 0.18738133, 0.2486899, 0.309017, 0.36812457, 0.4257793,
            0.4817537, 0.53582686,
        ];
        let gain = buffer.gain();
        let expected = 0.0102376845;
        assert_eq!(gain, expected);
    }

    #[test]
    fn difference_test() {
        let mut buffer = vec![
            0.0, 0.5, 1.0, 0.5, 0.0, -0.5, -1.0, -0.5, 0.0, 0.0, 0.5, 1.0, 0.5, 0.0, -0.5, -1.0,
            -0.5, 0.0, 0.5, 1.0, 0.5, 0.0, -0.5, -1.0, -0.5, 0.0, 0.5, 1.0, 0.5, 0.0, -0.5, -1.0,
        ];
        let expected = vec![
            0.0, 4.25, 11.5, 16.75, 22.0, 20.25, 13.5, 6.75, 2.0, 1.75, 7.75, 15.75, 21.75, 21.75,
            15.75, 7.75,
        ];
        buffer.yin_difference();
        assert_eq!(buffer, expected);
    }

    #[test]
    fn cumulative_mean_normalized_difference_test() {
        let mut buffer = vec![
            0.0, 4.25, 11.5, 16.75, 22.0, 20.25, 13.5, 6.75, 2.0, 1.75, 7.75, 15.75, 21.75, 21.75,
            15.75, 7.75,
        ];
        let expected = [
            0.0, 1.0, 1.4603175, 1.5461539, 1.6146789, 1.354515, 0.91784704, 0.4973684, 0.16494845,
            0.15949367, 0.7276996, 1.4171779, 1.8125, 1.7058824, 1.214876, 0.6142668,
        ];
        buffer.yin_cumulative_mean_normalized_difference();
        assert_eq!(buffer, expected);
    }

    #[test]
    fn absolute_threshold_test() {
        let mut buffer = vec![
            0.0, 1.0, 1.4603175, 1.5461539, 1.6146789, 1.354515, 0.91784704, 0.4973684, 0.16494845,
            0.15949367, 0.7276996, 1.4171779, 1.8125, 1.7058824, 1.214876, 0.614266,
        ];
        let threshold = 0.2;
        let expected = Some(9);
        assert_eq!(buffer.yin_absolute_threshold(threshold), expected);
    }

    #[test]
    fn parabolic_interpolation_test() {
        let mut buffer = vec![
            0.0, 1.0, 1.4603175, 1.5461539, 1.6146789, 1.354515, 0.91784704, 0.4973684, 0.16494845,
            0.15949367, 0.7276996, 1.4171779, 1.8125, 1.7058824, 1.214876, 0.6142668,
        ];
        let tau_estimate = 9;
        let expected = 8.509509;

        assert_eq!(buffer.yin_parabolic_interpolation(tau_estimate), expected);
    }

    #[test]
    fn yin_end_to_end() {
        let sample_rate = 44_100.00;
        let threshold = 0.20;

        let mut buffer = vec![
            0.0, 0.5, 1.0, 0.5, 0.0, -0.5, -1.0, -0.5, 0.0, 0.0, 0.5, 1.0, 0.5, 0.0, -0.5, -1.0,
            -0.5, 0.0, 0.5, 1.0, 0.5, 0.0, -0.5, -1.0, -0.5, 0.0, 0.5, 1.0, 0.5, 0.0, -0.5, -1.0,
        ];
        let expected = (5181.9604, 0.8405063);

        assert_eq!(buffer.yin_pitch_detection(sample_rate, threshold), expected);
    }
}