0

我正在尝试在 C++ 中计算 Mfcc 功能。我找到了 Aubio(https://github.com/aubio/aubio),但我不能产生与 Python 的 Librosa 相同的结果(这很重要)。Librosa 代码:

X, sample_rate = sf.read(file_name, dtype='float32')
mfccs = librosa.feature.mfcc(y=X, sr=sample_rate, n_mfcc=40)

奥比奥代码:

#include "utils.h"
#include "parse_args.h"
#include <stdlib.h>

aubio_pvoc_t *pv;    // a phase vocoder
cvec_t *fftgrain;    // outputs a spectrum
aubio_mfcc_t * mfcc; // which the mfcc will process
fvec_t * mfcc_out;   // to get the output coefficients

uint_t n_filters = 128;
uint_t n_coefs = 40;

 void process_block (fvec_t *ibuf, fvec_t *obuf)
 {
  fvec_zeros(obuf);
  //compute mag spectrum
  aubio_pvoc_do (pv, ibuf, fftgrain);
  //compute mfccs
  aubio_mfcc_do(mfcc, fftgrain, mfcc_out);

 }

 void process_print (void)
{
  /* output times in selected format */
  print_time (blocks * hop_size);
  outmsg ("\t");
  /* output extracted mfcc */
  fvec_print (mfcc_out);
}

int main(int argc, char **argv) {
  int ret = 0;
  // change some default params
  buffer_size  = 2048;
  hop_size = 512;

  examples_common_init(argc,argv);

  verbmsg ("using source: %s at %dHz\n", source_uri, samplerate);
  verbmsg ("buffer_size: %d, ", buffer_size);
  verbmsg ("hop_size: %d\n", hop_size);

  pv = new_aubio_pvoc (buffer_size, hop_size);
  fftgrain = new_cvec (buffer_size);
  mfcc = new_aubio_mfcc(buffer_size, n_filters, n_coefs, samplerate);
  mfcc_out = new_fvec(n_coefs);
  if (pv == NULL || fftgrain == NULL || mfcc == NULL || mfcc_out == NULL) {
    ret = 1;
    goto beach;
  }
  examples_common_process(process_block, process_print);
  printf("\nlen=%u\n", mfcc_out->length);
  del_aubio_pvoc (pv);
  del_cvec (fftgrain);
  del_aubio_mfcc(mfcc);
  del_fvec(mfcc_out);

beach:
  examples_common_del();
  return ret;
}

请帮助获得与 Librosa 相同的结果,或者建议任何 C++ 库都可以做到这一点。谢谢

4

1 回答 1

1

这可能是您正在寻找的内容:C Speech Features 该库是python_speech_features到 C 的完整端口,根据文档,您应该能够在 C++ 项目中使用它。结果与 Librosa 不同,这就是原因,但您应该能够使用它们。

于 2020-03-04T14:56:57.373 回答