static void SpectralEnvelopeEstimation(double *x, int x_length,
WorldParameters *world_parameters) {
CheapTrickOption option = {0};
InitializeCheapTrickOption(&option);
// This value may be better one for HMM speech synthesis.
// Default value is -0.09.
option.q1 = -0.15;
// Important notice (2016/02/02)
// You can control a parameter used for the lowest F0 in speech.
// You must not set the f0_floor to 0.
// It will cause a fatal error because fft_size indicates the infinity.
// You must not change the f0_floor after memory allocation.
// You should check the fft_size before excucing the analysis/synthesis.
// The default value (71.0) is strongly recommended.
// On the other hand, setting the lowest F0 of speech is a good choice
// to reduce the fft_size.
option.f0_floor = 71.0;
// Parameters setting and memory allocation.
world_parameters->fft_size =
GetFFTSizeForCheapTrick(world_parameters->fs, &option);
world_parameters->spectrogram = (double **) malloc(sizeof(double *) * (world_parameters->f0_length));
for (int i = 0; i < world_parameters->f0_length; ++i) {
world_parameters->spectrogram[i] =
(double*) malloc(sizeof(double) * (world_parameters->fft_size / 2 + 1));
}
DWORD elapsed_time = timeGetTime();
CheapTrick(x, x_length, world_parameters->fs, world_parameters->time_axis,
world_parameters->f0, world_parameters->f0_length, &option,
world_parameters->spectrogram);
printf("CheapTrick: %d [msec]\n", timeGetTime() - elapsed_time);
}
void InitializeCheapTrickOption(int fs, CheapTrickOption *option) {
// q1 is the parameter used for the spectral recovery.
// Since The parameter is optimized, you don't need to change the parameter.
option->q1 = -0.15;
// f0_floor and fs are used to determine fft_size;
// We strongly recommend not to change this value unless you have enough
// knowledge of the signal processing in CheapTrick.
option->f0_floor = world::kFloorF0;
option->fft_size = GetFFTSizeForCheapTrick(fs, option);
}
void CheapTrick(double *x, int x_length, int fs, double *time_axis, double *f0,
int f0_length, double **spectrogram) {
int fft_size = GetFFTSizeForCheapTrick(fs);
double *spectral_envelope = new double[fft_size];
ForwardRealFFT forward_real_fft = {0};
InitializeForwardRealFFT(fft_size, &forward_real_fft);
InverseRealFFT inverse_real_fft = {0};
InitializeInverseRealFFT(fft_size, &inverse_real_fft);
double current_f0;
for (int i = 0; i < f0_length; ++i) {
current_f0 = f0[i] <= world::kFloorF0 ? world::kDefaultF0 : f0[i];
CheapTrickGeneralBody(x, x_length, fs, current_f0, fft_size,
time_axis[i], &forward_real_fft, &inverse_real_fft, spectral_envelope);
for (int j = 0; j <= fft_size / 2; ++j)
spectrogram[i][j] = spectral_envelope[j];
}
DestroyForwardRealFFT(&forward_real_fft);
DestroyInverseRealFFT(&inverse_real_fft);
delete[] spectral_envelope;
}
//-----------------------------------------------------------------------------
// Test program.
// test.exe input.wav outout.wav f0 spec flag
// input.wav : argv[1] Input file
// output.wav : argv[2] Output file
// f0 : argv[3] F0 scaling (a positive number)
// spec : argv[4] Formant shift (a positive number)
//-----------------------------------------------------------------------------
int main(int argc, char *argv[]) {
if (argc != 2 && argc != 3 && argc != 4 && argc != 5) {
printf("error\n");
return 0;
}
int fs, nbit, x_length;
double *x = wavread(argv[1], &fs, &nbit, &x_length);
if (CheckLoadedFile(x, fs, nbit, x_length) == false) {
printf("error: File not found.\n");
return 0;
}
// Allocate memories
// The number of samples for F0
int f0_length = GetSamplesForDIO(fs, x_length, FRAMEPERIOD);
double *f0 = new double[f0_length];
double *time_axis = new double[f0_length];
// FFT size for CheapTrick
int fft_size = GetFFTSizeForCheapTrick(fs);
double **spectrogram = new double *[f0_length];
double **aperiodicity = new double *[f0_length];
for (int i = 0; i < f0_length; ++i) {
spectrogram[i] = new double[fft_size / 2 + 1];
aperiodicity[i] = new double[fft_size / 2 + 1];
}
// F0 estimation
F0Estimation(x, x_length, fs, f0_length, f0, time_axis);
// Spectral envelope estimation
SpectralEnvelopeEstimation(x, x_length, fs, time_axis, f0, f0_length,
spectrogram);
// Aperiodicity estimation by D4C
AperiodicityEstimation(x, x_length, fs, time_axis, f0, f0_length,
fft_size, aperiodicity);
// Note that F0 must not be changed until all parameters are estimated.
ParameterModification(argc, argv, fs, f0, f0_length, spectrogram);
// The length of the output waveform
int y_length =
static_cast<int>((f0_length - 1) * FRAMEPERIOD / 1000.0 * fs) + 1;
double *y = new double[y_length];
// Synthesis
WaveformSynthesis(f0, f0_length, spectrogram, aperiodicity, fft_size,
FRAMEPERIOD, fs, y_length, y);
// Output
wavwrite(y, y_length, fs, 16, argv[2]);
printf("complete.\n");
delete[] x;
delete[] time_axis;
delete[] f0;
delete[] y;
for (int i = 0; i < f0_length; ++i) {
delete[] spectrogram[i];
delete[] aperiodicity[i];
}
delete[] spectrogram;
delete[] aperiodicity;
return 0;
}
