10 #include <linux/soundcard.h>
12 #define SAMPLE_RATE 22050
13 #define FFT_LENGTH 4096 /* in samples */
14 #define PAD_FACTOR 2 /* 1/pf of the FFT samples are real samples, the rest are padding */
15 #define OVERLAP 4 /* 1/ol samples will be replaced in the buffer every frame. Should be
16 * a multiple of 2 for the Hamming window (see
17 * http://www-ccrma.stanford.edu/~jos/parshl/Choice_Hop_Size.html).
20 #define EQUAL_TEMPERAMENT 0
21 #define WELL_TEMPERED_GUITAR 1
23 #define TUNING WELL_TEMPERED_GUITAR
26 void read_chunk(int fd, double *in, unsigned num_samples);
27 void apply_window(double *in, double *out, unsigned num_samples);
28 std::pair<double, double> find_peak(double *in, unsigned num_samples);
29 void find_peak_magnitudes(std::complex<double> *in, double *out, unsigned num_samples);
30 double bin_to_freq(double bin, unsigned num_samples);
31 std::string freq_to_tonename(double freq);
32 std::pair<double, double> interpolate_peak(double ym1, double y0, double y1);
33 void print_spectrogram(double freq, double amp);
34 void write_sine(int dsp_fd, double freq, unsigned num_samples);
38 double *in, *in_windowed;
39 std::complex<double> *out;
44 in = reinterpret_cast<double *> (fftw_malloc(sizeof(double) * FFT_LENGTH / PAD_FACTOR));
45 in_windowed = reinterpret_cast<double *> (fftw_malloc(sizeof(double) * FFT_LENGTH));
46 out = reinterpret_cast<std::complex<double> *> (fftw_malloc(sizeof(std::complex<double>) * (FFT_LENGTH / 2 + 1)));
47 bins = reinterpret_cast<double *> (fftw_malloc(sizeof(double) * FFT_LENGTH / 2 + 1));
49 memset(in, 0, sizeof(double) * FFT_LENGTH / PAD_FACTOR);
52 p = fftw_plan_dft_r2c_1d(FFT_LENGTH, in_windowed, reinterpret_cast<fftw_complex *> (out), FFTW_ESTIMATE);
54 int fd = get_dsp_fd();
56 read_chunk(fd, in, FFT_LENGTH);
57 apply_window(in, in_windowed, FFT_LENGTH);
59 find_peak_magnitudes(out, bins, FFT_LENGTH);
60 std::pair<double, double> peak = find_peak(bins, FFT_LENGTH);
62 if (peak.first < 50.0 || peak.second - log10(FFT_LENGTH) < 0.0) {
63 #if TUNING == WELL_TEMPERED_GUITAR
66 printf("............\n");
69 print_spectrogram(peak.first, peak.second - log10(FFT_LENGTH));
76 int fd = open("/dev/dsp", O_RDWR);
82 ioctl(3, SNDCTL_DSP_RESET, 0);
84 int fmt = AFMT_S16_LE; // FIXME
85 ioctl(fd, SNDCTL_DSP_SETFMT, &fmt);
88 ioctl(fd, SOUND_PCM_WRITE_CHANNELS, &chan);
90 int rate = SAMPLE_RATE;
91 ioctl(fd, SOUND_PCM_WRITE_RATE, &rate);
93 ioctl(3, SNDCTL_DSP_SYNC, 0);
99 void read_chunk(int fd, double *in, unsigned num_samples)
102 unsigned to_read = num_samples / PAD_FACTOR / OVERLAP;
105 memmove(in, in + to_read, (num_samples / PAD_FACTOR - to_read) * sizeof(double));
107 ret = read(fd, buf, to_read * sizeof(short));
113 if (ret != int(to_read * sizeof(short))) {
119 for (unsigned i = 0; i < to_read; ++i)
120 in[i + (num_samples / PAD_FACTOR - to_read)] = double(buf[i]);
123 // make a pure 440hz sine for testing
124 void read_chunk(int fd, double *in, unsigned num_samples)
126 static double theta = 0.0;
127 for (unsigned i = 0; i < num_samples; ++i) {
129 theta += 2.0 * M_PI * 440.0 / double(SAMPLE_RATE);
134 void write_sine(int dsp_fd, double freq, unsigned num_samples)
136 static double theta = 0.0;
137 short buf[num_samples];
139 for (unsigned i = 0; i < num_samples; ++i) {
140 buf[i] = short(cos(theta) * 16384.0);
141 theta += 2.0 * M_PI * freq / double(SAMPLE_RATE);
144 write(dsp_fd, buf, num_samples * sizeof(short));
147 // Apply a standard Hamming window to our input data.
148 void apply_window(double *in, double *out, unsigned num_samples)
150 static double *win_data;
151 static unsigned win_len;
152 static bool win_inited = false;
154 // Initialize the window for the first time
156 win_len = num_samples / PAD_FACTOR;
157 win_data = new double[win_len];
159 for (unsigned i = 0; i < win_len; ++i) {
160 win_data[i] = 0.54 - 0.46 * cos(2.0 * M_PI * double(i) / double(win_len - 1));
166 assert(win_len == num_samples / PAD_FACTOR);
168 for (unsigned i = 0; i < win_len; ++i) {
169 out[i] = in[i] * win_data[i];
171 for (unsigned i = win_len; i < num_samples; ++i) {
176 void find_peak_magnitudes(std::complex<double> *in, double *out, unsigned num_samples)
178 for (unsigned i = 0; i < num_samples / 2 + 1; ++i)
182 std::pair<double, double> find_peak(double *in, unsigned num_samples)
184 double best_peak = in[0];
185 unsigned best_bin = 0;
187 for (unsigned i = 1; i < num_samples / 2 + 1; ++i) {
188 if (in[i] > best_peak) {
194 if (best_bin == 0 || best_bin == num_samples / 2) {
195 return std::make_pair(-1.0, 0.0);
199 printf("undertone strength: %+4.2f %+4.2f %+4.2f [%+4.2f] %+4.2f %+4.2f %+4.2f\n",
200 20.0 * log10(in[best_bin*4] / FFT_LENGTH),
201 20.0 * log10(in[best_bin*3] / FFT_LENGTH),
202 20.0 * log10(in[best_bin*2] / FFT_LENGTH),
203 20.0 * log10(in[best_bin] / FFT_LENGTH),
204 20.0 * log10(in[best_bin/2] / FFT_LENGTH),
205 20.0 * log10(in[best_bin/3] / FFT_LENGTH),
206 20.0 * log10(in[best_bin/4] / FFT_LENGTH));
209 // see if we might have hit an overtone (set a limit of 5dB)
210 for (unsigned i = 4; i >= 1; --i) {
211 if (best_bin != best_bin / i &&
212 20.0 * log10(in[best_bin] / in[best_bin / i]) < 5.0f) {
214 printf("Overtone of degree %u!\n", i);
218 // consider sliding one bin up or down
219 if (best_bin > 0 && in[best_bin - 1] > in[best_bin] && in[best_bin - 1] > in[best_bin - 2]) {
221 } else if (best_bin < num_samples / 2 && in[best_bin + 1] > in[best_bin] && in[best_bin + 1] > in[best_bin + 2]) {
229 std::pair<double, double> peak =
230 interpolate_peak(in[best_bin - 1],
234 return std::make_pair(bin_to_freq(double(best_bin) + peak.first, num_samples), peak.second);
237 double bin_to_freq(double bin, unsigned num_samples)
239 return bin * SAMPLE_RATE / double(num_samples);
243 * Given three bins, find the interpolated real peak based
244 * on their magnitudes. To do this, we execute the following
247 * Fit a polynomial of the form ax^2 + bx + c = 0 to the data
248 * we have. Maple readily yields our coefficients, assuming
249 * that we have the values at x=-1, x=0 and x=1:
251 * > f := x -> a*x^2 + b*x + c;
254 * f := x -> a x + b x + c
256 * > cf := solve({ f(-1) = ym1, f(0) = y0, f(1) = y1 }, { a, b, c });
259 * cf := {c = y0, b = ---- - ---, a = ---- + --- - y0}
262 * Now let's find the maximum point for the polynomial (it has to be
263 * a maximum, since y0 is the greatest value):
265 * > xmax := solve(subs(cf, diff(f(x), x)) = 0, x);
268 * xmax := -------------------
269 * 2 (y1 + ym1 - 2 y0)
271 * We could go further and insert {fmax,a,b,c} into the original
272 * polynomial, but it just gets hairy. We calculate a, b and c separately
275 * http://www-ccrma.stanford.edu/~jos/parshl/Peak_Detection_Steps_3.html
276 * claims that detection is almost twice as good when using dB scale instead
277 * of linear scale for the input values, so we use that. (As a tiny bonus,
278 * we get back dB scale from the function.)
280 std::pair<double, double> interpolate_peak(double ym1, double y0, double y1)
291 double a = 0.5 * y1 + 0.5 * ym1 - y0;
292 double b = 0.5 * y1 - 0.5 * ym1;
295 double xmax = (ym1 - y1) / (2.0 * (y1 + ym1 - 2.0 * y0));
296 double ymax = 20.0 * (a * xmax * xmax + b * xmax + c) - 90.0;
298 return std::make_pair(xmax, ymax);
301 std::string freq_to_tonename(double freq)
303 std::string notenames[] = { "C", "C#", "D", "D#", "E", "F", "F#", "G", "G#", "A", "A#", "B" };
304 double half_notes_away = 12.0 * log2(freq / 440.0) - 3.0;
305 int hnai = int(floor(half_notes_away + 0.5));
306 int octave = (hnai + 48) / 12;
309 sprintf(buf, "%s%d + %.2f [%d]", notenames[((hnai % 12) + 12) % 12].c_str(), octave, half_notes_away - hnai, hnai);
313 #if TUNING == EQUAL_TEMPERAMENT
314 void print_spectrogram(double freq, double amp)
316 std::string notenames[] = { "C", "C#", "D", "D#", "E", "F", "F#", "G", "G#", "A", "A#", "B" };
317 double half_notes_away = 12.0 * log2(freq / 440.0) - 3.0;
318 int hnai = int(floor(half_notes_away + 0.5));
319 int octave = (hnai + 48) / 12;
321 for (int i = 0; i < 12; ++i)
322 if (i == ((hnai % 12) + 12) % 12)
327 printf(" (%-2s%d %+.2f, %5.2fHz) [%5.2fdB] [", notenames[((hnai % 12) + 12) % 12].c_str(), octave, half_notes_away - hnai,
330 double off = half_notes_away - hnai;
331 for (int i = -10; i <= 10; ++i) {
332 if (off >= (i-0.5) * 0.05 && off < (i+0.5) * 0.05) {
350 static note notes[] = {
351 { "E-3", 110.0 * (3.0/4.0) },
353 { "D-4", 110.0 * (4.0/3.0) },
354 { "G-4", 110.0 * (4.0/3.0)*(4.0/3.0) },
355 { "B-4", 440.0 * (3.0/4.0)*(3.0/4.0) },
356 { "E-5", 440.0 * (3.0/4.0) }
359 void print_spectrogram(double freq, double amp)
361 double best_away = 999999999.9;
362 unsigned best_away_ind = 0;
364 for (unsigned i = 0; i < sizeof(notes)/sizeof(note); ++i) {
365 double half_notes_away = 12.0 * log2(freq / notes[i].freq);
366 if (fabs(half_notes_away) < fabs(best_away)) {
367 best_away = half_notes_away;
372 for (unsigned i = 0; i < sizeof(notes)/sizeof(note); ++i)
373 if (i == best_away_ind)
378 printf(" (%s %+.2f, %5.2fHz) [%5.2fdB] [", notes[best_away_ind].notename, best_away, freq, amp);
380 for (int i = -10; i <= 10; ++i) {
381 if (best_away >= (i-0.5) * 0.05 && best_away < (i+0.5) * 0.05) {