void
aubio_beattracking_do (aubio_beattracking_t * bt, const fvec_t * dfframe,
fvec_t * output)
{
uint_t i, k;
uint_t step = bt->step;
uint_t laglen = bt->rwv->length;
uint_t winlen = bt->dfwv->length;
uint_t maxindex = 0;
//number of harmonics in shift invariant comb filterbank
uint_t numelem = 4;
smpl_t phase; // beat alignment (step - lastbeat)
smpl_t beat; // beat position
smpl_t bp; // beat period
uint_t a, b; // used to build shift invariant comb filterbank
uint_t kmax; // number of elements used to find beat phase
/* copy dfframe, apply detection function weighting, and revert */
fvec_copy (dfframe, bt->dfrev);
fvec_weight (bt->dfrev, bt->dfwv);
fvec_rev (bt->dfrev);
/* compute autocorrelation function */
aubio_autocorr (dfframe, bt->acf);
/* if timesig is unknown, use metrically unbiased version of filterbank */
if (!bt->timesig) {
numelem = 4;
} else {
numelem = bt->timesig;
}
/* first and last output values are left intentionally as zero */
fvec_zeros (bt->acfout);
/* compute shift invariant comb filterbank */
for (i = 1; i < laglen - 1; i++) {
for (a = 1; a <= numelem; a++) {
for (b = 1; b < 2 * a; b++) {
bt->acfout->data[i] += bt->acf->data[i * a + b - 1]
* 1. / (2. * a - 1.);
}
}
}
/* apply Rayleigh weight */
fvec_weight (bt->acfout, bt->rwv);
/* find non-zero Rayleigh period */
maxindex = fvec_max_elem (bt->acfout);
if (maxindex > 0 && maxindex < bt->acfout->length - 1) {
bt->rp = fvec_quadratic_peak_pos (bt->acfout, maxindex);
} else {
bt->rp = bt->rayparam;
}
/* activate biased filterbank */
aubio_beattracking_checkstate (bt);
#if 0 // debug metronome mode
bt->bp = 36.9142;
#endif
bp = bt->bp;
/* end of biased filterbank */
if (bp == 0) {
fvec_zeros(output);
return;
}
/* deliberate integer operation, could be set to 3 max eventually */
kmax = FLOOR (winlen / bp);
/* initialize output */
fvec_zeros (bt->phout);
for (i = 0; i < bp; i++) {
for (k = 0; k < kmax; k++) {
bt->phout->data[i] += bt->dfrev->data[i + (uint_t) ROUND (bp * k)];
}
}
fvec_weight (bt->phout, bt->phwv);
/* find Rayleigh period */
maxindex = fvec_max_elem (bt->phout);
if (maxindex >= winlen - 1) {
#if AUBIO_BEAT_WARNINGS
AUBIO_WRN ("no idea what this groove's phase is\n");
#endif /* AUBIO_BEAT_WARNINGS */
phase = step - bt->lastbeat;
} else {
phase = fvec_quadratic_peak_pos (bt->phout, maxindex);
}
/* take back one frame delay */
phase += 1.;
#if 0 // debug metronome mode
phase = step - bt->lastbeat;
#endif
/* reset output */
fvec_zeros (output);
i = 1;
beat = bp - phase;
// AUBIO_DBG ("bp: %f, phase: %f, lastbeat: %f, step: %d, winlen: %d\n",
// bp, phase, bt->lastbeat, step, winlen);
/* the next beat will be earlier than 60% of the tempo period
skip this one */
if ( ( step - bt->lastbeat - phase ) < -0.40 * bp ) {
#if AUBIO_BEAT_WARNINGS
AUBIO_WRN ("back off-beat error, skipping this beat\n");
#endif /* AUBIO_BEAT_WARNINGS */
beat += bp;
}
/* start counting the beats */
while (beat + bp < 0) {
beat += bp;
}
if (beat >= 0) {
//AUBIO_DBG ("beat: %d, %f, %f\n", i, bp, beat);
output->data[i] = beat;
i++;
}
while (beat + bp <= step) {
beat += bp;
//AUBIO_DBG ("beat: %d, %f, %f\n", i, bp, beat);
output->data[i] = beat;
i++;
}
bt->lastbeat = beat;
/* store the number of beats in this frame as the first element */
output->data[0] = i;
}
aubio_source_avcodec_t * new_aubio_source_avcodec(char_t * path, uint_t samplerate, uint_t hop_size) {
aubio_source_avcodec_t * s = AUBIO_NEW(aubio_source_avcodec_t);
int err;
if (path == NULL) {
AUBIO_ERR("Aborted opening null path\n");
goto beach;
}
if ((sint_t)samplerate < 0) {
AUBIO_ERR("Can not open %s with samplerate %d\n", path, samplerate);
goto beach;
}
if ((sint_t)hop_size <= 0) {
AUBIO_ERR("Can not open %s with hop_size %d\n", path, hop_size);
goto beach;
}
s->hop_size = hop_size;
s->channels = 1;
s->path = path;
// register all formats and codecs
av_register_all();
// if path[0] != '/'
//avformat_network_init();
// try opening the file and get some info about it
AVFormatContext *avFormatCtx = s->avFormatCtx;
avFormatCtx = NULL;
if ( (err = avformat_open_input(&avFormatCtx, s->path, NULL, NULL) ) < 0 ) {
char errorstr[256];
av_strerror (err, errorstr, sizeof(errorstr));
AUBIO_ERR("Failed opening %s (%s)\n", s->path, errorstr);
goto beach;
}
// try to make sure max_analyze_duration is big enough for most songs
avFormatCtx->max_analyze_duration *= 100;
// retrieve stream information
if ( (err = avformat_find_stream_info(avFormatCtx, NULL)) < 0 ) {
char errorstr[256];
av_strerror (err, errorstr, sizeof(errorstr));
AUBIO_ERR("Could not find stream information " "for %s (%s)\n", s->path,
errorstr);
goto beach;
}
// dump information about file onto standard error
//av_dump_format(avFormatCtx, 0, s->path, 0);
// look for the first audio stream
uint_t i;
sint_t selected_stream = -1;
for (i = 0; i < avFormatCtx->nb_streams; i++) {
if (avFormatCtx->streams[i]->codec->codec_type == AVMEDIA_TYPE_AUDIO) {
if (selected_stream == -1) {
selected_stream = i;
} else {
AUBIO_WRN("More than one audio stream in %s, "
"taking the first one\n", s->path);
}
}
}
if (selected_stream == -1) {
AUBIO_ERR("No audio stream in %s\n", s->path);
goto beach;
}
//AUBIO_DBG("Taking stream %d in file %s\n", selected_stream, s->path);
s->selected_stream = selected_stream;
AVCodecContext *avCodecCtx = s->avCodecCtx;
avCodecCtx = avFormatCtx->streams[selected_stream]->codec;
AVCodec *codec = avcodec_find_decoder(avCodecCtx->codec_id);
if (codec == NULL) {
AUBIO_ERR("Could not find decoder for %s", s->path);
goto beach;
}
if ( ( err = avcodec_open2(avCodecCtx, codec, NULL) ) < 0) {
char errorstr[256];
av_strerror (err, errorstr, sizeof(errorstr));
AUBIO_ERR("Could not load codec for %s (%s)\n", s->path, errorstr);
goto beach;
}
/* get input specs */
s->input_samplerate = avCodecCtx->sample_rate;
s->input_channels = avCodecCtx->channels;
//AUBIO_DBG("input_samplerate: %d\n", s->input_samplerate);
//AUBIO_DBG("input_channels: %d\n", s->input_channels);
if (samplerate == 0) {
samplerate = s->input_samplerate;
//AUBIO_DBG("sampling rate set to 0, automagically adjusting to %d\n", samplerate);
}
s->samplerate = samplerate;
if (s->samplerate > s->input_samplerate) {
AUBIO_WRN("upsampling %s from %d to %d\n", s->path,
s->input_samplerate, s->samplerate);
}
AVFrame *avFrame = s->avFrame;
avFrame = avcodec_alloc_frame();
if (!avFrame) {
AUBIO_ERR("Could not allocate frame for (%s)\n", s->path);
}
/* allocate output for avr */
s->output = (float *)av_malloc(AUBIO_AVCODEC_MAX_BUFFER_SIZE * sizeof(float));
s->read_samples = 0;
s->read_index = 0;
s->avFormatCtx = avFormatCtx;
s->avCodecCtx = avCodecCtx;
s->avFrame = avFrame;
// default to mono output
aubio_source_avcodec_reset_resampler(s, 0);
s->eof = 0;
s->multi = 0;
//av_log_set_level(AV_LOG_QUIET);
return s;
beach:
//AUBIO_ERR("can not read %s at samplerate %dHz with a hop_size of %d\n",
// s->path, s->samplerate, s->hop_size);
del_aubio_source_avcodec(s);
return NULL;
}
void
aubio_beattracking_checkstate (aubio_beattracking_t * bt)
{
uint_t i, j, a, b;
uint_t flagconst = 0;
sint_t counter = bt->counter;
uint_t flagstep = bt->flagstep;
smpl_t gp = bt->gp;
smpl_t bp = bt->bp;
smpl_t rp = bt->rp;
smpl_t rp1 = bt->rp1;
smpl_t rp2 = bt->rp2;
uint_t laglen = bt->rwv->length;
uint_t acflen = bt->acf->length;
uint_t step = bt->step;
fvec_t *acf = bt->acf;
fvec_t *acfout = bt->acfout;
if (gp) {
// compute shift invariant comb filterbank
fvec_zeros (acfout);
for (i = 1; i < laglen - 1; i++) {
for (a = 1; a <= bt->timesig; a++) {
for (b = 1; b < 2 * a; b++) {
acfout->data[i] += acf->data[i * a + b - 1];
}
}
}
// since gp is set, gwv has been computed in previous checkstate
fvec_weight (acfout, bt->gwv);
gp = fvec_quadratic_peak_pos (acfout, fvec_max_elem (acfout));
} else {
//still only using general model
gp = 0;
}
//now look for step change - i.e. a difference between gp and rp that
// is greater than 2*constthresh - always true in first case, since gp = 0
if (counter == 0) {
if (ABS (gp - rp) > 2. * bt->g_var) {
flagstep = 1; // have observed step change.
counter = 3; // setup 3 frame counter
} else {
flagstep = 0;
}
}
//i.e. 3rd frame after flagstep initially set
if (counter == 1 && flagstep == 1) {
//check for consistency between previous beatperiod values
if (ABS (2 * rp - rp1 - rp2) < bt->g_var) {
//if true, can activate context dependent model
flagconst = 1;
counter = 0; // reset counter and flagstep
} else {
//if not consistent, then don't flag consistency!
flagconst = 0;
counter = 2; // let it look next time
}
} else if (counter > 0) {
//if counter doesn't = 1,
counter = counter - 1;
}
rp2 = rp1;
rp1 = rp;
if (flagconst) {
/* first run of new hypothesis */
gp = rp;
bt->timesig = fvec_gettimesig (acf, acflen, gp);
for (j = 0; j < laglen; j++)
bt->gwv->data[j] =
EXP (-.5 * SQR ((smpl_t) (j + 1. - gp)) / SQR (bt->g_var));
flagconst = 0;
bp = gp;
/* flat phase weighting */
fvec_ones (bt->phwv);
} else if (bt->timesig) {
/* context dependant model */
bp = gp;
/* gaussian phase weighting */
if (step > bt->lastbeat) {
for (j = 0; j < 2 * laglen; j++) {
bt->phwv->data[j] =
EXP (-.5 * SQR ((smpl_t) (1. + j - step +
bt->lastbeat)) / (bp / 8.));
}
} else {
//AUBIO_DBG("NOT using phase weighting as step is %d and lastbeat %d \n",
// step,bt->lastbeat);
fvec_ones (bt->phwv);
}
} else {
/* initial state */
bp = rp;
/* flat phase weighting */
fvec_ones (bt->phwv);
}
/* do some further checks on the final bp value */
/* if tempo is > 206 bpm, half it */
while (0 < bp && bp < 25) {
#if AUBIO_BEAT_WARNINGS
AUBIO_WRN ("doubling from %f (%f bpm) to %f (%f bpm)\n",
bp, 60.*44100./512./bp, bp/2., 60.*44100./512./bp/2. );
//AUBIO_DBG("warning, halving the tempo from %f\n", 60.*samplerate/hopsize/bp);
#endif /* AUBIO_BEAT_WARNINGS */
bp = bp * 2;
}
//AUBIO_DBG("tempo:\t%3.5f bpm | ", 5168./bp);
/* smoothing */
//bp = (uint_t) (0.8 * (smpl_t)bp + 0.2 * (smpl_t)bp2);
//AUBIO_DBG("tempo:\t%3.5f bpm smoothed | bp2 %d | bp %d | ", 5168./bp, bp2, bp);
//bp2 = bp;
//AUBIO_DBG("time signature: %d \n", bt->timesig);
bt->counter = counter;
bt->flagstep = flagstep;
bt->gp = gp;
bt->bp = bp;
bt->rp1 = rp1;
bt->rp2 = rp2;
}
int aubio_audio_unit_blocking(aubio_audio_unit_t *o)
{
uint_t sw_output_channels, sw_input_channels,
hw_output_channels, hw_input_channels,
i, j, blocksize;
if (! o->callback) return -1;
smpl_t ** tbuf;
sw_output_channels = o->sw_output_channels;
sw_input_channels = o->sw_input_channels;
hw_output_channels = o->hw_output_channels;
hw_input_channels = o->hw_input_channels;
blocksize = o->blocksize;
if (!sw_input_channels && !sw_output_channels) goto fail;
if (o->dio_error) {
AUBIO_WRN("audio_unit: dio error %d\n", o->total_frames);
o->dio_error = 0;
}
if (o->au_ios_inbuf) {
/* copy samples from input buffer */
tbuf = o->input_frames->data;
if (o->input_enabled) {
for (j = 0; j < blocksize;j++) {
for (i = 0; i < sw_input_channels && i < hw_input_channels; i++) {
//tbuf[i][j] =
// (smpl_t)(o->au_ios_inbuf[i + (j + o->au_ios_end) * sw_input_channels] / 32768.);
// on iphone, input is mono, copy right to left channel
tbuf[i][j] =
(smpl_t) o->au_ios_inbuf[0 + (j + o->au_ios_end) * hw_input_channels]
* INT_TO_FLOAT;
}
}
} else {
// input is disabled, fill with zeroes
for (j = 0; j < blocksize; j++) {
for (i = 0; i < sw_input_channels && i < hw_input_channels; i++) {
tbuf[i][j] = 0;
}
}
}
}
o->callback(o->callback_closure, o->input_frames, o->output_frames);
/* copy samples to output buffer */
tbuf = o->output_frames->data;
for (i = 0; i < o->output_frames->height; i++) {
for (j = 0; j < o->output_frames->length; j++) {
smpl_t val = tbuf[i][j];
if (val < -1.0) val = -1.0;
if (val > 1.0) val = 1.0;
o->au_ios_outbuf[i + (j + o->au_ios_end) * hw_output_channels ] = (SInt16)(val * 32767);
}
}
o->au_ios_end += blocksize;
o->au_ios_end %= AU_IOS_MAX_FRAMES;
o->total_frames += blocksize;
return 0;
fail:
AUBIO_ERR("audio_unit: callback() failed\n");
o->total_frames += AU_IOS_MAX_FRAMES;
return 1;
}
aubio_source_apple_audio_t * new_aubio_source_apple_audio(char_t * path, uint_t samplerate, uint_t block_size)
{
aubio_source_apple_audio_t * s = AUBIO_NEW(aubio_source_apple_audio_t);
s->path = path;
s->block_size = block_size;
OSStatus err = noErr;
UInt32 propSize;
// open the resource url
CFURLRef fileURL = getURLFromPath(path);
err = ExtAudioFileOpenURL(fileURL, &s->audioFile);
if (err) { AUBIO_ERR("error when trying to access %s, in ExtAudioFileOpenURL, %d\n", s->path, (int)err); goto beach;}
// create an empty AudioStreamBasicDescription
AudioStreamBasicDescription fileFormat;
propSize = sizeof(fileFormat);
memset(&fileFormat, 0, sizeof(AudioStreamBasicDescription));
// fill it with the file description
err = ExtAudioFileGetProperty(s->audioFile,
kExtAudioFileProperty_FileDataFormat, &propSize, &fileFormat);
if (err) { AUBIO_ERROR("error in ExtAudioFileGetProperty, %d\n", (int)err); goto beach;}
if (samplerate == 0) {
samplerate = fileFormat.mSampleRate;
//AUBIO_DBG("sampling rate set to 0, automagically adjusting to %d\n", samplerate);
}
s->samplerate = samplerate;
s->source_samplerate = fileFormat.mSampleRate;
s->channels = fileFormat.mChannelsPerFrame;
AudioStreamBasicDescription clientFormat;
propSize = sizeof(clientFormat);
memset(&clientFormat, 0, sizeof(AudioStreamBasicDescription));
clientFormat.mFormatID = kAudioFormatLinearPCM;
clientFormat.mSampleRate = (Float64)(s->samplerate);
clientFormat.mFormatFlags = kAudioFormatFlagIsSignedInteger | kAudioFormatFlagIsPacked;
clientFormat.mChannelsPerFrame = s->channels;
clientFormat.mBitsPerChannel = sizeof(short) * 8;
clientFormat.mFramesPerPacket = 1;
clientFormat.mBytesPerFrame = clientFormat.mBitsPerChannel * clientFormat.mChannelsPerFrame / 8;
clientFormat.mBytesPerPacket = clientFormat.mFramesPerPacket * clientFormat.mBytesPerFrame;
clientFormat.mReserved = 0;
// set the client format description
err = ExtAudioFileSetProperty(s->audioFile, kExtAudioFileProperty_ClientDataFormat,
propSize, &clientFormat);
if (err) {
AUBIO_ERROR("error in ExtAudioFileSetProperty, %d\n", (int)err);
#if 1
// print client and format descriptions
AUBIO_DBG("Opened %s\n", s->path);
AUBIO_DBG("file/client Format.mFormatID: : %3c%c%c%c / %c%c%c%c\n",
(int)RT_BYTE4(fileFormat.mFormatID), (int)RT_BYTE3(fileFormat.mFormatID), (int)RT_BYTE2(fileFormat.mFormatID), (int)RT_BYTE1(fileFormat.mFormatID),
(int)RT_BYTE4(clientFormat.mFormatID), (int)RT_BYTE3(clientFormat.mFormatID), (int)RT_BYTE2(clientFormat.mFormatID), (int)RT_BYTE1(clientFormat.mFormatID)
);
AUBIO_DBG("file/client Format.mSampleRate : %6.0f / %.0f\n", fileFormat.mSampleRate , clientFormat.mSampleRate);
AUBIO_DBG("file/client Format.mFormatFlags : %6d / %d\n", (int)fileFormat.mFormatFlags , (int)clientFormat.mFormatFlags);
AUBIO_DBG("file/client Format.mChannelsPerFrame : %6d / %d\n", (int)fileFormat.mChannelsPerFrame, (int)clientFormat.mChannelsPerFrame);
AUBIO_DBG("file/client Format.mBitsPerChannel : %6d / %d\n", (int)fileFormat.mBitsPerChannel , (int)clientFormat.mBitsPerChannel);
AUBIO_DBG("file/client Format.mFramesPerPacket : %6d / %d\n", (int)fileFormat.mFramesPerPacket , (int)clientFormat.mFramesPerPacket);
AUBIO_DBG("file/client Format.mBytesPerFrame : %6d / %d\n", (int)fileFormat.mBytesPerFrame , (int)clientFormat.mBytesPerFrame);
AUBIO_DBG("file/client Format.mBytesPerPacket : %6d / %d\n", (int)fileFormat.mBytesPerPacket , (int)clientFormat.mBytesPerPacket);
AUBIO_DBG("file/client Format.mReserved : %6d / %d\n", (int)fileFormat.mReserved , (int)clientFormat.mReserved);
#endif
goto beach;
}
// compute the size of the segments needed to read the input file
UInt32 samples = s->block_size * clientFormat.mChannelsPerFrame;
Float64 rateRatio = clientFormat.mSampleRate / fileFormat.mSampleRate;
uint_t segmentSize= (uint_t)(samples * rateRatio + .5);
if (rateRatio < 1.) {
segmentSize = (uint_t)(samples / rateRatio + .5);
} else if (rateRatio > 1.) {
AUBIO_WRN("up-sampling %s from %0.2fHz to %0.2fHz\n", s->path, fileFormat.mSampleRate, clientFormat.mSampleRate);
} else {
assert ( segmentSize == samples );
//AUBIO_DBG("not resampling, segmentSize %d, block_size %d\n", segmentSize, s->block_size);
}
// allocate the AudioBufferList
if (createAubioBufferList(&s->bufferList, s->channels, segmentSize)) err = -1;
return s;
beach:
AUBIO_FREE(s);
return NULL;
}
