Changeset 2eb52bd for src/spectral

Timestamp:

Nov 17, 2018, 4:36:09 PM (6 years ago)

Author:

Paul Brossier <piem@piem.org>

Branches:

feature/autosink, feature/cnn, feature/cnn_org, feature/constantq, feature/crepe, feature/crepe_org, feature/pitchshift, feature/pydocstrings, feature/timestretch, fix/ffmpeg5, master

Children:

62c2d00, a95b386

Parents:

adde1ba (diff), 8eecb9f (diff)
Note: this is a merge changeset, the changes displayed below correspond to the merge itself.
Use the (diff) links above to see all the changes relative to each parent.

Message:

Merge branch 'feature/mfccparams'

Location:

src/spectral

Files:

• filterbank_mel.h (modified) (1 diff)
• mfcc.c (modified) (6 diffs)
• mfcc.h (modified) (1 diff)

src/spectral/filterbank_mel.h

@@ -56 +56 @@
 
   \param fb filterbank object
-  \param samplerate audio sampling rate
+  \param samplerate audio sampling rate, in Hz
 
   The filter coefficients are built to match exactly Malcolm Slaney's Auditory

src/spectral/mfcc.c

@@ -52 +52 @@
   fvec_t *output;
 #endif
+  smpl_t scale;
 };
 
@@ -75 +76 @@
   /* filterbank allocation */
   mfcc->fb = new_aubio_filterbank (n_filters, mfcc->win_s);
-  aubio_filterbank_set_mel_coeffs_slaney (mfcc->fb, samplerate);
+  if (n_filters == 40)
+    aubio_filterbank_set_mel_coeffs_slaney (mfcc->fb, samplerate);
+  else
+    aubio_filterbank_set_mel_coeffs(mfcc->fb, samplerate,
+        0, samplerate/2.);
 
   /* allocating buffers */
@@ -97 +102 @@
   mfcc->output = new_fvec (n_filters);
 #endif
+
+  mfcc->scale = 1.;
 
   return mfcc;
@@ -128 +135 @@
   fvec_t tmp;
 #endif
+
   /* compute filterbank */
   aubio_filterbank_do (mf->fb, in, mf->in_dct);
@@ -134 +142 @@
   fvec_log10 (mf->in_dct);
 
-  /* raise power */
-  //fvec_pow (mf->in_dct, 3.);
+  if (mf->scale != 1) fvec_mul (mf->in_dct, mf->scale);
 
   /* compute mfccs */
@@ -151 +158 @@
   return;
 }
+
+uint_t aubio_mfcc_set_power (aubio_mfcc_t *mf, smpl_t power)
+{
+  return aubio_filterbank_set_power(mf->fb, power);
+}
+
+uint_t aubio_mfcc_get_power (aubio_mfcc_t *mf)
+{
+  return aubio_filterbank_get_power(mf->fb);
+}
+
+uint_t aubio_mfcc_set_scale (aubio_mfcc_t *mf, smpl_t scale)
+{
+  mf->scale = scale;
+  return AUBIO_OK;
+}
+
+uint_t aubio_mfcc_get_scale (aubio_mfcc_t *mf)
+{
+  return mf->scale;
+}
+
+uint_t aubio_mfcc_set_mel_coeffs (aubio_mfcc_t *mf, smpl_t freq_min,
+    smpl_t freq_max)
+{
+  return aubio_filterbank_set_mel_coeffs(mf->fb, mf->samplerate,
+      freq_min, freq_max);
+}
+
+uint_t aubio_mfcc_set_mel_coeffs_htk (aubio_mfcc_t *mf, smpl_t freq_min,
+    smpl_t freq_max)
+{
+  return aubio_filterbank_set_mel_coeffs_htk(mf->fb, mf->samplerate,
+      freq_min, freq_max);
+}
+
+uint_t aubio_mfcc_set_mel_coeffs_slaney (aubio_mfcc_t *mf)
+{
+  return aubio_filterbank_set_mel_coeffs_slaney (mf->fb, mf->samplerate);
+}

src/spectral/mfcc.h

@@ -74 +74 @@
 void aubio_mfcc_do (aubio_mfcc_t * mf, const cvec_t * in, fvec_t * out);
 
+/** set power parameter
+
+  \param mf mfcc object, as returned by new_aubio_mfcc()
+  \param power Raise norm of the input spectrum norm to this power before
+  computing filterbank.  Defaults to `1`.
+
+  See aubio_filterbank_set_power().
+
+ */
+uint_t aubio_mfcc_set_power (aubio_mfcc_t *mf, smpl_t power);
+
+/** get power parameter
+
+  \param mf mfcc object, as returned by new_aubio_mfcc()
+  \return current power parameter. Defaults to `1`.
+
+  See aubio_filterbank_get_power().
+
+ */
+uint_t aubio_mfcc_get_power (aubio_mfcc_t *mf);
+
+/** set scaling parameter
+
+  \param mf mfcc object, as returned by new_aubio_mfcc()
+  \param scale Scaling value to apply.
+
+  Scales the output of the filterbank after taking its logarithm and before
+  computing the DCT. Defaults to `1`.
+
+*/
+uint_t aubio_mfcc_set_scale (aubio_mfcc_t *mf, smpl_t scale);
+
+/** get scaling parameter
+
+  \param mf mfcc object, as returned by new_aubio_mfcc()
+  \return current scaling parameter. Defaults to `1`.
+
+ */
+uint_t aubio_mfcc_get_scale (aubio_mfcc_t *mf);
+
+/** Mel filterbank initialization
+
+  \param mf mfcc object
+  \param fmin start frequency, in Hz
+  \param fmax end frequency, in Hz
+
+  The filterbank will be initialized with bands linearly spaced in the mel
+  scale, from `fmin` to `fmax`.
+
+  See also
+  --------
+
+  aubio_filterbank_set_mel_coeffs()
+
+*/
+uint_t aubio_mfcc_set_mel_coeffs (aubio_mfcc_t *mf,
+        smpl_t fmin, smpl_t fmax);
+
+/** Mel filterbank initialization
+
+  \param mf mfcc object
+  \param fmin start frequency, in Hz
+  \param fmax end frequency, in Hz
+
+  The bank of filters will be initalized to to cover linearly spaced bands in
+  the Htk mel scale, from `fmin` to `fmax`.
+
+  See also
+  --------
+
+  aubio_filterbank_set_mel_coeffs_htk()
+
+*/
+uint_t aubio_mfcc_set_mel_coeffs_htk (aubio_mfcc_t *mf,
+        smpl_t fmin, smpl_t fmax);
+
+/** Mel filterbank initialization (Auditory Toolbox's parameters)
+
+  \param mf mfcc object
+  \param samplerate audio sampling rate, in Hz
+
+  The filter coefficients are built to match exactly Malcolm Slaney's Auditory
+  Toolbox implementation. The number of filters should be 40.
+
+  This is the default filterbank when `mf` was created with `n_filters = 40`.
+
+  See also
+  --------
+
+  aubio_filterbank_set_mel_coeffs_slaney()
+
+*/
+uint_t aubio_mfcc_set_mel_coeffs_slaney (aubio_mfcc_t *mf);
+
 #ifdef __cplusplus
 }