ref: 350211958fe957d85c19fad553f88327bf39d5d2
parent: 0eca01fde43b53a98ede876a6bf7ec9785ca9ae3
parent: c82859f347b082d73b66648b10070a4c45434b9b
author: Paul Brossier <piem@piem.org>
date: Wed Apr 5 07:49:57 EDT 2017
Merge branch 'awhitening'
--- a/examples/aubioonset.c
+++ b/examples/aubioonset.c
@@ -43,10 +43,11 @@
} else {aubio_wavetable_stop ( wavetable );
}
- if (mix_input)
+ if (mix_input) {aubio_wavetable_do (wavetable, ibuf, obuf);
- else
+ } else {aubio_wavetable_do (wavetable, obuf, obuf);
+ }
}
void process_print (void)
@@ -61,13 +62,6 @@
int ret = 0;
examples_common_init(argc,argv);
- verbmsg ("using source: %s at %dHz\n", source_uri, samplerate);- verbmsg ("onset method: %s, ", onset_method);- verbmsg ("buffer_size: %d, ", buffer_size);- verbmsg ("hop_size: %d, ", hop_size);- verbmsg ("silence: %f, ", silence_threshold);- verbmsg ("threshold: %f\n", onset_threshold);-
o = new_aubio_onset (onset_method, buffer_size, hop_size, samplerate);
if (o == NULL) { ret = 1; goto beach; }if (onset_threshold != 0.)
@@ -76,6 +70,15 @@
aubio_onset_set_silence (o, silence_threshold);
if (onset_minioi != 0.)
aubio_onset_set_minioi_s (o, onset_minioi);
+
+ verbmsg ("using source: %s at %dHz\n", source_uri, samplerate);+ verbmsg ("onset method: %s, ", onset_method);+ verbmsg ("buffer_size: %d, ", buffer_size);+ verbmsg ("hop_size: %d, ", hop_size);+ verbmsg ("silence: %f, ", aubio_onset_get_silence(o));+ verbmsg ("threshold: %f, ", aubio_onset_get_threshold(o));+ verbmsg ("awhitening: %f, ", aubio_onset_get_awhitening(o));+ verbmsg ("compression: %f\n", aubio_onset_get_compression(o));onset = new_fvec (1);
--- a/examples/aubiotrack.c
+++ b/examples/aubiotrack.c
@@ -46,10 +46,11 @@
} else {aubio_wavetable_stop ( wavetable );
}
- if (mix_input)
+ if (mix_input) {aubio_wavetable_do (wavetable, ibuf, obuf);
- else
+ } else {aubio_wavetable_do (wavetable, obuf, obuf);
+ }
}
void process_print (void) {--- a/python/lib/gen_code.py
+++ b/python/lib/gen_code.py
@@ -183,17 +183,21 @@
def gen_code(self):
out = ""
- out += self.gen_struct()
- out += self.gen_doc()
- out += self.gen_new()
- out += self.gen_init()
- out += self.gen_del()
- out += self.gen_do()
- out += self.gen_memberdef()
- out += self.gen_set()
- out += self.gen_get()
- out += self.gen_methodef()
- out += self.gen_typeobject()
+ try:
+ out += self.gen_struct()
+ out += self.gen_doc()
+ out += self.gen_new()
+ out += self.gen_init()
+ out += self.gen_del()
+ out += self.gen_do()
+ out += self.gen_memberdef()
+ out += self.gen_set()
+ out += self.gen_get()
+ out += self.gen_methodef()
+ out += self.gen_typeobject()
+ except Exception as e:
+ print ("Failed generating code for", self.shortname)+ raise
return out
def gen_struct(self):
--- a/python/lib/gen_external.py
+++ b/python/lib/gen_external.py
@@ -39,6 +39,7 @@
'source_wavread',
#'sampler',
'audio_unit',
+ 'spectral_whitening',
]
def get_preprocessor():
--- a/python/tests/test_onset.py
+++ b/python/tests/test_onset.py
@@ -19,25 +19,25 @@
self.o = onset(samplerate = self.samplerate)
def test_get_delay(self):
- assert_equal (self.o.get_delay(), int(4.3 * self.o.hop_size))
+ self.assertGreater(self.o.get_delay(), 0)
def test_get_delay_s(self):
- assert_almost_equal (self.o.get_delay_s(), self.o.get_delay() / float(self.samplerate))
+ self.assertGreater(self.o.get_delay_s(), 0.)
def test_get_delay_ms(self):
- assert_almost_equal (self.o.get_delay_ms(), self.o.get_delay() * 1000. / self.samplerate, 5)
+ self.assertGreater(self.o.get_delay_ms(), 0.)
def test_get_minioi(self):
- assert_almost_equal (self.o.get_minioi(), 0.02 * self.samplerate)
+ self.assertGreater(self.o.get_minioi(), 0)
def test_get_minioi_s(self):
- assert_almost_equal (self.o.get_minioi_s(), 0.02)
+ self.assertGreater(self.o.get_minioi_s(), 0.)
def test_get_minioi_ms(self):
- assert_equal (self.o.get_minioi_ms(), 20.)
+ self.assertGreater(self.o.get_minioi_ms(), 0.)
def test_get_threshold(self):
- assert_almost_equal (self.o.get_threshold(), 0.3)
+ self.assertGreater(self.o.get_threshold(), 0.)
def test_set_delay(self):
val = 256
--- a/src/aubio.h
+++ b/src/aubio.h
@@ -187,6 +187,7 @@
#include "spectral/filterbank_mel.h"
#include "spectral/mfcc.h"
#include "spectral/specdesc.h"
+#include "spectral/awhitening.h"
#include "spectral/tss.h"
#include "pitch/pitch.h"
#include "onset/onset.h"
--- a/src/cvec.c
+++ b/src/cvec.c
@@ -139,3 +139,10 @@
cvec_norm_zeros(s);
cvec_phas_zeros(s);
}
+
+void cvec_logmag(cvec_t *s, smpl_t lambda) {+ uint_t j;
+ for (j=0; j< s->length; j++) {+ s->norm[j] = LOG(lambda * s->norm[j] + 1);
+ }
+}
--- a/src/cvec.h
+++ b/src/cvec.h
@@ -230,6 +230,16 @@
*/
void cvec_zeros(cvec_t *s);
+/** take logarithmic magnitude
+
+ \param s input cvec to compress
+ \param lambda value to use for normalisation
+
+ \f$ S_k = log( \lambda * S_k + 1 ) \f$
+
+*/
+void cvec_logmag(cvec_t *s, smpl_t lambda);
+
#ifdef __cplusplus
}
#endif
--- a/src/mathutils.c
+++ b/src/mathutils.c
@@ -289,6 +289,25 @@
}
}
+void fvec_push(fvec_t *in, smpl_t new_elem) {+ uint_t i;
+ for (i = 0; i < in->length - 1; i++) {+ in->data[i] = in->data[i + 1];
+ }
+ in->data[in->length - 1] = new_elem;
+}
+
+void fvec_clamp(fvec_t *in, smpl_t absmax) {+ uint_t i;
+ for (i = 0; i < in->length; i++) {+ if (in->data[i] > 0 && in->data[i] > ABS(absmax)) {+ in->data[i] = absmax;
+ } else if (in->data[i] < 0 && in->data[i] < -ABS(absmax)) {+ in->data[i] = -absmax;
+ }
+ }
+}
+
smpl_t
aubio_level_lin (const fvec_t * f)
{--- a/src/mathutils.h
+++ b/src/mathutils.h
@@ -117,6 +117,17 @@
*/
void fvec_ishift (fvec_t * v);
+/** push a new element to the end of a vector, erasing the first element and
+ * sliding all others
+
+ \param in vector to push to
+ \param new_elem new_element to add at the end of the vector
+
+ In numpy words, this is equivalent to: in = np.concatenate([in, [new_elem]])[1:]
+
+*/
+void fvec_push(fvec_t *in, smpl_t new_elem);
+
/** compute the sum of all elements of a vector
\param v vector to compute the sum of
--- a/src/musicutils.h
+++ b/src/musicutils.h
@@ -156,6 +156,14 @@
*/
smpl_t aubio_level_detection (const fvec_t * v, smpl_t threshold);
+/** clamp the values of a vector within the range [-abs(max), abs(max)]
+
+ \param in vector to clamp
+ \param absmax maximum value over which input vector elements should be clamped
+
+*/
+void fvec_clamp(fvec_t *in, smpl_t absmax);
+
#ifdef __cplusplus
}
#endif
--- a/src/onset/onset.c
+++ b/src/onset/onset.c
@@ -23,10 +23,13 @@
#include "cvec.h"
#include "spectral/specdesc.h"
#include "spectral/phasevoc.h"
+#include "spectral/awhitening.h"
#include "onset/peakpicker.h"
#include "mathutils.h"
#include "onset/onset.h"
+void aubio_onset_default_parameters (aubio_onset_t *o, const char_t * method);
+
/** structure to store object state */
struct _aubio_onset_t {aubio_pvoc_t * pv; /**< phase vocoder */
@@ -42,6 +45,11 @@
uint_t total_frames; /**< total number of frames processed since the beginning */
uint_t last_onset; /**< last detected onset location, in frames */
+
+ uint_t apply_compression;
+ smpl_t lambda_compression;
+ uint_t apply_awhitening; /**< apply adaptive spectral whitening */
+ aubio_spectral_whitening_t *spectral_whitening;
};
/* execute onset detection function on iput buffer */
@@ -49,6 +57,16 @@
{smpl_t isonset = 0;
aubio_pvoc_do (o->pv,input, o->fftgrain);
+ /*
+ if (apply_filtering) {+ }
+ */
+ if (o->apply_awhitening) {+ aubio_spectral_whitening_do(o->spectral_whitening, o->fftgrain);
+ }
+ if (o->apply_compression) {+ cvec_logmag(o->fftgrain, o->lambda_compression);
+ }
aubio_specdesc_do (o->od, o->fftgrain, o->desc);
aubio_peakpicker_do(o->pp, o->desc, onset);
isonset = onset->data[0];
@@ -57,10 +75,17 @@
//AUBIO_DBG ("silent onset, not marking as onset\n");isonset = 0;
} else {+ // we have an onset
uint_t new_onset = o->total_frames + (uint_t)ROUND(isonset * o->hop_size);
+ // check if last onset time was more than minioi ago
if (o->last_onset + o->minioi < new_onset) {- //AUBIO_DBG ("accepted detection, marking as onset\n");- o->last_onset = new_onset;
+ // start of file: make sure (new_onset - delay) >= 0
+ if (o->last_onset > 0 && o->delay > new_onset) {+ isonset = 0;
+ } else {+ //AUBIO_DBG ("accepted detection, marking as onset\n");+ o->last_onset = MAX(o->delay, new_onset);
+ }
} else { //AUBIO_DBG ("doubled onset, not marking as onset\n");isonset = 0;
@@ -99,6 +124,32 @@
return aubio_onset_get_last_s (o) * 1000.;
}
+uint_t aubio_onset_set_awhitening (aubio_onset_t *o, uint_t enable)
+{+ o->apply_awhitening = enable == 1 ? 1 : 0;
+ return AUBIO_OK;
+}
+
+smpl_t aubio_onset_get_awhitening (aubio_onset_t *o)
+{+ return o->apply_awhitening;
+}
+
+uint_t aubio_onset_set_compression (aubio_onset_t *o, smpl_t lambda)
+{+ if (lambda < 0.) {+ return AUBIO_FAIL;
+ }
+ o->lambda_compression = lambda;
+ o->apply_compression = (o->lambda_compression > 0.) ? 1 : 0;
+ return AUBIO_OK;
+}
+
+smpl_t aubio_onset_get_compression (aubio_onset_t *o)
+{+ return o->apply_compression ? o->lambda_compression : 0;
+}
+
uint_t aubio_onset_set_silence(aubio_onset_t * o, smpl_t silence) {o->silence = silence;
return AUBIO_OK;
@@ -208,16 +259,12 @@
if (o->od == NULL) goto beach_specdesc;
o->fftgrain = new_cvec(buf_size);
o->desc = new_fvec(1);
+ o->spectral_whitening = new_aubio_spectral_whitening(buf_size, hop_size, samplerate);
- /* set some default parameter */
- aubio_onset_set_threshold (o, 0.3);
- aubio_onset_set_delay(o, 4.3 * hop_size);
- aubio_onset_set_minioi_ms(o, 20.);
- aubio_onset_set_silence(o, -70.);
-
/* initialize internal variables */
- o->last_onset = 0;
- o->total_frames = 0;
+ aubio_onset_set_default_parameters (o, onset_mode);
+
+ aubio_onset_reset(o);
return o;
beach_specdesc:
@@ -228,8 +275,69 @@
return NULL;
}
+void aubio_onset_reset (aubio_onset_t *o) {+ o->last_onset = 0;
+ o->total_frames = 0;
+}
+
+uint_t aubio_onset_set_default_parameters (aubio_onset_t * o, const char_t * onset_mode)
+{+ uint_t ret = AUBIO_OK;
+ /* set some default parameter */
+ aubio_onset_set_threshold (o, 0.3);
+ aubio_onset_set_delay (o, 4.3 * o->hop_size);
+ aubio_onset_set_minioi_ms (o, 50.);
+ aubio_onset_set_silence (o, -70.);
+ // disable spectral whitening
+ aubio_onset_set_awhitening (o, 0);
+ // disable logarithmic magnitude
+ aubio_onset_set_compression (o, 0.);
+
+ /* method specific optimisations */
+ if (strcmp (onset_mode, "energy") == 0) {+ } else if (strcmp (onset_mode, "hfc") == 0 || strcmp (onset_mode, "default") == 0) {+ aubio_onset_set_threshold (o, 0.058);
+ aubio_onset_set_compression (o, 1.);
+ } else if (strcmp (onset_mode, "complexdomain") == 0
+ || strcmp (onset_mode, "complex") == 0) {+ aubio_onset_set_delay (o, 4.6 * o->hop_size);
+ aubio_onset_set_threshold (o, 0.15);
+ aubio_onset_set_awhitening(o, 1);
+ aubio_onset_set_compression (o, 1.);
+ } else if (strcmp (onset_mode, "phase") == 0) {+ o->apply_compression = 0;
+ aubio_onset_set_awhitening (o, 0);
+ } else if (strcmp (onset_mode, "mkl") == 0) {+ aubio_onset_set_threshold (o, 0.05);
+ aubio_onset_set_awhitening(o, 1);
+ aubio_onset_set_compression (o, 0.02);
+ } else if (strcmp (onset_mode, "kl") == 0) {+ aubio_onset_set_threshold (o, 0.35);
+ aubio_onset_set_awhitening(o, 1);
+ aubio_onset_set_compression (o, 0.02);
+ } else if (strcmp (onset_mode, "specflux") == 0) {+ aubio_onset_set_threshold (o, 0.18);
+ aubio_onset_set_awhitening(o, 1);
+ aubio_spectral_whitening_set_relax_time(o->spectral_whitening, 100);
+ aubio_spectral_whitening_set_floor(o->spectral_whitening, 1.);
+ aubio_onset_set_compression (o, 10.);
+ } else if (strcmp (onset_mode, "specdiff") == 0) {+ } else if (strcmp (onset_mode, "old_default") == 0) {+ // used to reproduce results obtained with the previous version
+ aubio_onset_set_threshold (o, 0.3);
+ aubio_onset_set_minioi_ms (o, 20.);
+ aubio_onset_set_compression (o, 0.);
+ } else {+ AUBIO_WRN("onset: unknown spectral descriptor type %s, "+ "using default parameters.\n", onset_mode);
+ ret = AUBIO_FAIL;
+ }
+ return ret;
+}
+
void del_aubio_onset (aubio_onset_t *o)
{+ del_aubio_spectral_whitening(o->spectral_whitening);
del_aubio_specdesc(o->od);
del_aubio_peakpicker(o->pp);
del_aubio_pvoc(o->pv);
--- a/src/onset/onset.h
+++ b/src/onset/onset.h
@@ -117,6 +117,44 @@
*/
smpl_t aubio_onset_get_last_ms (const aubio_onset_t *o);
+/** set onset detection adaptive whitening
+
+ \param o onset detection object as returned by new_aubio_onset()
+ \param enable 1 to enable, 0 to disable
+
+ \return 0 if successful, 1 otherwise
+
+*/
+uint_t aubio_onset_set_awhitening(aubio_onset_t * o, uint_t enable);
+
+/** get onset detection adaptive whitening
+
+ \param o onset detection object as returned by new_aubio_onset()
+
+ \return 1 if enabled, 0 otherwise
+
+*/
+smpl_t aubio_onset_get_awhitening(aubio_onset_t * o);
+
+/** set or disable log compression
+
+ \param o onset detection object as returned by new_aubio_onset()
+ \param lambda logarithmic compression factor, 0 to disable
+
+ \return 0 if successful, 1 otherwise
+
+ */
+uint_t aubio_onset_set_compression(aubio_onset_t *o, smpl_t lambda);
+
+/** get onset detection log compression
+
+ \param o onset detection object as returned by new_aubio_onset()
+
+ \returns 0 if disabled, compression factor otherwise
+
+ */
+smpl_t aubio_onset_get_compression(aubio_onset_t *o);
+
/** set onset detection silence threshold
\param o onset detection object as returned by new_aubio_onset()
@@ -273,6 +311,27 @@
*/
smpl_t aubio_onset_get_threshold(const aubio_onset_t * o);
+
+/** set default parameters
+
+ \param o onset detection object as returned by new_aubio_onset()
+ \param onset_mode detection mode to adjust
+
+ This function is called at the end of new_aubio_onset().
+
+ */
+uint_t aubio_onset_set_default_parameters (aubio_onset_t * o, const char_t * onset_mode);
+
+/** reset onset detection
+
+ \param o onset detection object as returned by new_aubio_onset()
+
+ Reset current time and last onset to 0.
+
+ This function is called at the end of new_aubio_onset().
+
+ */
+void aubio_onset_reset(aubio_onset_t * o);
/** delete onset detection object
--- a/src/onset/peakpicker.c
+++ b/src/onset/peakpicker.c
@@ -92,27 +92,21 @@
fvec_t *thresholded = p->thresholded;
fvec_t *scratch = p->scratch;
smpl_t mean = 0., median = 0.;
- uint_t length = p->win_post + p->win_pre + 1;
uint_t j = 0;
- /* store onset in onset_keep */
- /* shift all elements but last, then write last */
- for (j = 0; j < length - 1; j++) {- onset_keep->data[j] = onset_keep->data[j + 1];
- onset_proc->data[j] = onset_keep->data[j];
- }
- onset_keep->data[length - 1] = onset->data[0];
- onset_proc->data[length - 1] = onset->data[0];
+ /* push new novelty to the end */
+ fvec_push(onset_keep, onset->data[0]);
+ /* store a copy */
+ fvec_copy(onset_keep, onset_proc);
- /* filter onset_proc */
- /** \bug filtfilt calculated post+pre times, should be only once !? */
+ /* filter this copy */
aubio_filter_do_filtfilt (p->biquad, onset_proc, scratch);
/* calculate mean and median for onset_proc */
mean = fvec_mean (onset_proc);
- /* copy to scratch */
- for (j = 0; j < length; j++)
- scratch->data[j] = onset_proc->data[j];
+
+ /* copy to scratch and compute its median */
+ fvec_copy(onset_proc, scratch);
median = p->thresholdfn (scratch);
/* shift peek array */
--- /dev/null
+++ b/src/spectral/awhitening.c
@@ -1,0 +1,120 @@
+/*
+ * Copyright (C) 2003-2015 Paul Brossier <piem@aubio.org>
+ *
+ * This file is part of aubio.
+ *
+ * aubio is free software: you can redistribute it and/or modify it under the
+ * terms of the GNU General Public License as published by the Free Software
+ * Foundation, either version 3 of the License, or (at your option) any later
+ * version.
+ *
+ * aubio is distributed in the hope that it will be useful, but WITHOUT ANY
+ * WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
+ * FOR A PARTICULAR PURPOSE. See the GNU General Public License for more
+ * details.
+ *
+ * You should have received a copy of the GNU General Public License along with
+ * aubio. If not, see <http://www.gnu.org/licenses/>.
+ *
+ */
+
+#include "aubio_priv.h"
+#include "fvec.h"
+#include "cvec.h"
+#include "mathutils.h"
+#include "spectral/awhitening.h"
+
+#define aubio_spectral_whitening_default_relax_time 250 // in seconds, between 22 and 446
+#define aubio_spectral_whitening_default_decay 0.001 // -60dB attenuation
+#define aubio_spectral_whitening_default_floor 1.e-4 // from 1.e-6 to .2
+
+/** structure to store object state */
+struct _aubio_spectral_whitening_t {+ uint_t buf_size;
+ uint_t hop_size;
+ uint_t samplerate;
+ smpl_t relax_time;
+ smpl_t r_decay;
+ smpl_t floor;
+ fvec_t *peak_values;
+};
+
+void
+aubio_spectral_whitening_do (aubio_spectral_whitening_t * o, cvec_t * fftgrain)
+{+ uint_t i = 0;
+ for (i = 0; i < o->peak_values->length; i++) {+ smpl_t tmp = MAX(o->r_decay * o->peak_values->data[i], o->floor);
+ o->peak_values->data[i] = MAX(fftgrain->norm[i], tmp);
+ fftgrain->norm[i] /= o->peak_values->data[i];
+ }
+}
+
+aubio_spectral_whitening_t *
+new_aubio_spectral_whitening (uint_t buf_size, uint_t hop_size, uint_t samplerate)
+{+ aubio_spectral_whitening_t *o = AUBIO_NEW (aubio_spectral_whitening_t);
+ if ((sint_t)buf_size < 1) {+ AUBIO_ERR("spectral_whitening: got buffer_size %d, but can not be < 1\n", buf_size);+ goto beach;
+ } else if ((sint_t)hop_size < 1) {+ AUBIO_ERR("spectral_whitening: got hop_size %d, but can not be < 1\n", hop_size);+ goto beach;
+ } else if ((sint_t)samplerate < 1) {+ AUBIO_ERR("spectral_whitening: got samplerate %d, but can not be < 1\n", samplerate);+ goto beach;
+ }
+ o->peak_values = new_fvec (buf_size / 2 + 1);
+ o->buf_size = buf_size;
+ o->hop_size = hop_size;
+ o->samplerate = samplerate;
+ o->floor = aubio_spectral_whitening_default_floor;
+ aubio_spectral_whitening_set_relax_time (o, aubio_spectral_whitening_default_relax_time);
+ aubio_spectral_whitening_reset (o);
+ return o;
+
+beach:
+ AUBIO_FREE(o);
+ return NULL;
+}
+
+uint_t
+aubio_spectral_whitening_set_relax_time (aubio_spectral_whitening_t * o, smpl_t relax_time)
+{+ o->relax_time = relax_time;
+ o->r_decay = POW (aubio_spectral_whitening_default_decay,
+ (o->hop_size / (float) o->samplerate) / o->relax_time);
+ return AUBIO_OK;
+}
+
+smpl_t
+aubio_spectral_whitening_get_relax_time (aubio_spectral_whitening_t * o)
+{+ return o->relax_time;
+}
+
+uint_t
+aubio_spectral_whitening_set_floor (aubio_spectral_whitening_t *o, smpl_t floor)
+{+ o->floor = floor;
+ return AUBIO_OK;
+}
+
+smpl_t aubio_spectral_whitening_get_floor (aubio_spectral_whitening_t *o)
+{+ return o->floor;
+}
+
+void
+aubio_spectral_whitening_reset (aubio_spectral_whitening_t * o)
+{+ /* cover the case n == 0. */
+ fvec_set_all (o->peak_values, o->floor);
+}
+
+void
+del_aubio_spectral_whitening (aubio_spectral_whitening_t * o)
+{+ del_fvec (o->peak_values);
+ AUBIO_FREE (o);
+}
--- /dev/null
+++ b/src/spectral/awhitening.h
@@ -1,0 +1,125 @@
+/*
+ Copyright (C) 2003-2015 Paul Brossier <piem@aubio.org>
+
+ This file is part of aubio.
+
+ aubio is free software: you can redistribute it and/or modify
+ it under the terms of the GNU General Public License as published by
+ the Free Software Foundation, either version 3 of the License, or
+ (at your option) any later version.
+
+ aubio is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ GNU General Public License for more details.
+
+ You should have received a copy of the GNU General Public License
+ along with aubio. If not, see <http://www.gnu.org/licenses/>.
+
+*/
+
+/** \file
+
+ Spectral adaptive whitening
+
+ References:
+
+ D. Stowell and M. D. Plumbley. Adaptive whitening for improved real-time
+ audio onset detection. In Proceedings of the International Computer Music
+ Conference (ICMC), 2007, Copenhagen, Denmark.
+
+ http://www.eecs.qmul.ac.uk/~markp/2007/StowellPlumbley07-icmc.pdf
+
+ S. Böck,, F. Krebs, and M. Schedl. Evaluating the Online Capabilities of
+ Onset Detection Methods. In Proceedings of the 13th International Society for
+ Music Information Retrieval Conference (ISMIR), 2012, Porto, Portugal.
+
+ http://ismir2012.ismir.net/event/papers/049_ISMIR_2012.pdf
+ http://www.cp.jku.at/research/papers/Boeck_etal_ISMIR_2012.pdf
+
+*/
+
+
+#ifndef _AUBIO_SPECTRAL_WHITENING_H
+#define _AUBIO_SPECTRAL_WHITENING_H
+
+#ifdef __cplusplus
+extern "C" {+#endif
+
+/** spectral whitening structure */
+typedef struct _aubio_spectral_whitening_t aubio_spectral_whitening_t;
+
+/** execute spectral adaptive whitening, in-place
+
+ \param o spectral whitening object as returned by new_aubio_spectral_whitening()
+ \param fftgrain input signal spectrum as computed by aubio_pvoc_do() or aubio_fft_do()
+
+*/
+void aubio_spectral_whitening_do (aubio_spectral_whitening_t * o,
+ cvec_t * fftgrain);
+
+/** creation of a spectral whitening object
+
+ \param buf_size window size of input grains
+ \param hop_size number of samples between two consecutive input grains
+ \param samplerate sampling rate of the input signal
+
+*/
+aubio_spectral_whitening_t *new_aubio_spectral_whitening (uint_t buf_size,
+ uint_t hop_size,
+ uint_t samplerate);
+
+/** reset spectral whitening object
+
+ \param o spectral whitening object as returned by new_aubio_spectral_whitening()
+
+ */
+void aubio_spectral_whitening_reset (aubio_spectral_whitening_t * o);
+
+/** set relaxation time for spectral whitening
+
+ \param o spectral whitening object as returned by new_aubio_spectral_whitening()
+ \param relax_time relaxation time in seconds between 20 and 500, defaults 250
+
+ */
+uint_t aubio_spectral_whitening_set_relax_time (aubio_spectral_whitening_t * o,
+ smpl_t relax_time);
+
+/** get relaxation time of spectral whitening
+
+ \param o spectral whitening object as returned by new_aubio_spectral_whitening()
+ \return relaxation time in seconds
+
+*/
+smpl_t aubio_spectral_whitening_get_relax_time (aubio_spectral_whitening_t * o);
+
+/** set floor for spectral whitening
+
+ \param o spectral whitening object as returned by new_aubio_spectral_whitening()
+ \param floor value (typically between 1.e-6 and .2, defaults to 1.e-4)
+
+ */
+uint_t aubio_spectral_whitening_set_floor (aubio_spectral_whitening_t * o,
+ smpl_t floor);
+
+/** get floor of spectral whitening
+
+ \param o spectral whitening object as returned by new_aubio_spectral_whitening()
+ \return floor value
+
+*/
+smpl_t aubio_spectral_whitening_get_floor (aubio_spectral_whitening_t * o);
+
+/** deletion of a spectral whitening
+
+ \param o spectral whitening object as returned by new_aubio_spectral_whitening()
+
+*/
+void del_aubio_spectral_whitening (aubio_spectral_whitening_t * o);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* _AUBIO_SPECTRAL_WHITENING_H */
--- a/src/spectral/specdesc.c
+++ b/src/spectral/specdesc.c
@@ -30,6 +30,7 @@
void aubio_specdesc_hfc(aubio_specdesc_t *o, const cvec_t * fftgrain, fvec_t * onset);
void aubio_specdesc_complex(aubio_specdesc_t *o, const cvec_t * fftgrain, fvec_t * onset);
void aubio_specdesc_phase(aubio_specdesc_t *o, const cvec_t * fftgrain, fvec_t * onset);
+void aubio_specdesc_wphase(aubio_specdesc_t *o, const cvec_t * fftgrain, fvec_t * onset);
void aubio_specdesc_specdiff(aubio_specdesc_t *o, const cvec_t * fftgrain, fvec_t * onset);
void aubio_specdesc_kl(aubio_specdesc_t *o, const cvec_t * fftgrain, fvec_t * onset);
void aubio_specdesc_mkl(aubio_specdesc_t *o, const cvec_t * fftgrain, fvec_t * onset);
@@ -57,6 +58,7 @@
aubio_onset_hfc, /**< high frequency content */
aubio_onset_complex, /**< complex domain */
aubio_onset_phase, /**< phase fast */
+ aubio_onset_wphase, /**< weighted phase */
aubio_onset_kl, /**< Kullback Liebler */
aubio_onset_mkl, /**< modified Kullback Liebler */
aubio_onset_specflux, /**< spectral flux */
@@ -159,6 +161,23 @@
//onset->data[0] = fvec_mean(o->dev1);
}
+/* weighted phase */
+void
+aubio_specdesc_wphase(aubio_specdesc_t *o,
+ const cvec_t *fftgrain, fvec_t *onset) {+ uint_t i;
+ aubio_specdesc_phase(o, fftgrain, onset);
+ for (i = 0; i < fftgrain->length; i++) {+ o->dev1->data[i] *= fftgrain->norm[i];
+ }
+ /* apply o->histogram */
+ aubio_hist_dyn_notnull(o->histog,o->dev1);
+ /* weight it */
+ aubio_hist_weight(o->histog);
+ /* its mean is the result */
+ onset->data[0] = aubio_hist_mean(o->histog);
+}
+
/* Spectral difference method onset detection function */
void aubio_specdesc_specdiff(aubio_specdesc_t *o,
const cvec_t * fftgrain, fvec_t * onset){@@ -250,6 +269,8 @@
onset_type = aubio_onset_complex;
else if (strcmp (onset_mode, "phase") == 0)
onset_type = aubio_onset_phase;
+ else if (strcmp (onset_mode, "wphase") == 0)
+ onset_type = aubio_onset_wphase;
else if (strcmp (onset_mode, "mkl") == 0)
onset_type = aubio_onset_mkl;
else if (strcmp (onset_mode, "kl") == 0)
@@ -270,6 +291,8 @@
onset_type = aubio_specmethod_decrease;
else if (strcmp (onset_mode, "rolloff") == 0)
onset_type = aubio_specmethod_rolloff;
+ else if (strcmp (onset_mode, "old_default") == 0)
+ onset_type = aubio_onset_default;
else if (strcmp (onset_mode, "default") == 0)
onset_type = aubio_onset_default;
else {@@ -291,6 +314,7 @@
o->theta2 = new_fvec(rsize);
break;
case aubio_onset_phase:
+ case aubio_onset_wphase:
o->dev1 = new_fvec(rsize);
o->theta1 = new_fvec(rsize);
o->theta2 = new_fvec(rsize);
@@ -325,6 +349,9 @@
case aubio_onset_phase:
o->funcpointer = aubio_specdesc_phase;
break;
+ case aubio_onset_wphase:
+ o->funcpointer = aubio_specdesc_wphase;
+ break;
case aubio_onset_specdiff:
o->funcpointer = aubio_specdesc_specdiff;
break;
@@ -378,6 +405,7 @@
del_fvec(o->theta2);
break;
case aubio_onset_phase:
+ case aubio_onset_wphase:
del_fvec(o->dev1);
del_fvec(o->theta1);
del_fvec(o->theta2);
--- a/src/spectral/specdesc.h
+++ b/src/spectral/specdesc.h
@@ -59,6 +59,13 @@
Conference on Acoustics Speech and Signal Processing, pages 441444,
Hong-Kong, 2003.
+ \b \p wphase : Weighted Phase Deviation onset detection function
+
+ S. Dixon. Onset detection revisited. In Proceedings of the 9th International
+ Conference on Digital Audio Ef- fects (DAFx) , pages 133–137, 2006.
+
+ http://www.eecs.qmul.ac.uk/~simond/pub/2006/dafx.pdf
+
\b \p specdiff : Spectral difference method onset detection function
Jonhatan Foote and Shingo Uchihashi. The beat spectrum: a new approach to
@@ -174,8 +181,11 @@
The parameter \p method is a string that can be any of:
- - `energy`, `hfc`, `complex`, `phase`, `specdiff`, `kl`, `mkl`, `specflux`
- - `centroid`, `spread`, `skewness`, `kurtosis`, `slope`, `decrease`, `rolloff`
+ - onset novelty functions: `complex`, `energy`, `hfc`, `kl`, `mkl`,
+ `phase`, `specdiff`, `specflux`, `wphase`,
+
+ - spectral descriptors: `centroid`, `decrease`, `kurtosis`, `rolloff`,
+ `skewness`, `slope`, `spread`.
*/
aubio_specdesc_t *new_aubio_specdesc (const char_t * method, uint_t buf_size);
--- a/src/synth/wavetable.c
+++ b/src/synth/wavetable.c
@@ -103,6 +103,7 @@
for (i = 0; i < output->length; i++) {output->data[i] += input->data[i];
}
+ fvec_clamp(output, 1.);
}
}
--- /dev/null
+++ b/tests/src/spectral/test-awhitening.c
@@ -1,0 +1,84 @@
+#include <aubio.h>
+#include "utils_tests.h"
+
+int main (int argc, char **argv)
+{+ sint_t err = 0;
+
+ if (argc < 3) {+ err = 2;
+ PRINT_ERR("not enough arguments\n");+ PRINT_MSG("usage: %s <input_path> <output_path> [samplerate] [hop_size]\n", argv[0]);+ return err;
+ }
+
+ uint_t samplerate = 0;
+ uint_t win_size = 1024;
+ uint_t hop_size = 512;
+ uint_t n_frames = 0, read = 0;
+
+ char_t *source_path = argv[1];
+ char_t *sink_path = argv[2];
+
+ if ( argc >= 4 ) samplerate = atoi(argv[3]);
+ if ( argc >= 5 ) hop_size = atoi(argv[4]);
+ if ( argc >= 6 ) {+ err = 2;
+ PRINT_ERR("too many arguments\n");+ return err;
+ }
+
+ fvec_t *vec = new_fvec(hop_size);
+ fvec_t *out = new_fvec(hop_size); // output buffer
+ fvec_t *scale = new_fvec(hop_size);
+ cvec_t *fftgrain = new_cvec(win_size); // fft norm and phase
+ if (!vec) { err = 1; goto beach_fvec; }+
+ aubio_source_t *i = new_aubio_source(source_path, samplerate, hop_size);
+ if (!i) { err = 1; goto beach_source; }+
+ if (samplerate == 0 ) samplerate = aubio_source_get_samplerate(i);
+
+ aubio_sink_t *o = new_aubio_sink(sink_path, samplerate);
+ if (!o) { err = 1; goto beach_sink; }+
+ aubio_pvoc_t *pv = new_aubio_pvoc(win_size, hop_size);
+
+ aubio_spectral_whitening_t *awhitening =
+ new_aubio_spectral_whitening (win_size, hop_size, samplerate);
+
+ aubio_spectral_whitening_set_relax_time(awhitening, 20.);
+ fvec_set_all(scale, 3.);
+
+ PRINT_MSG("spectral whitening relaxation time is %f\n",+ aubio_spectral_whitening_get_relax_time(awhitening));
+
+ do {+ aubio_source_do(i, vec, &read);
+ aubio_pvoc_do(pv, vec, fftgrain);
+ // apply spectral whitening
+ aubio_spectral_whitening_do(awhitening, fftgrain);
+ // rebuild the signal
+ aubio_pvoc_rdo(pv, fftgrain, out);
+ // make louder
+ fvec_weight(out, scale);
+ // make sure we dont saturate
+ fvec_clamp(out, 1.);
+ // write output
+ aubio_sink_do(o, out, read);
+ n_frames += read;
+ } while ( read == hop_size );
+
+ PRINT_MSG("read %d frames at %dHz (%d blocks) from %s written to %s\n",+ n_frames, samplerate, n_frames / hop_size,
+ source_path, sink_path);
+
+ del_aubio_sink(o);
+beach_sink:
+ del_aubio_source(i);
+beach_source:
+ del_fvec(vec);
+beach_fvec:
+ return err;
+}
+