// Get address of region(s) from which we can read data.
// If the region is contiguous, |data_ptr_bytes_2| will be zero.
// If non-contiguous, |data_ptr_bytes_2| will be the size in bytes of the second
// region. Returns room available to be read or |element_count|, whichever is
// smaller.
static size_t GetBufferReadRegions(RingBuffer* buf,
size_t element_count,
void** data_ptr_1,
size_t* data_ptr_bytes_1,
void** data_ptr_2,
size_t* data_ptr_bytes_2) {
const size_t readable_elements = WebRtc_available_read(buf);
const size_t read_elements = (readable_elements < element_count ?
readable_elements : element_count);
const size_t margin = buf->element_count - buf->read_pos;
// Check to see if read is not contiguous.
if (read_elements > margin) {
// Write data in two blocks that wrap the buffer.
*data_ptr_1 = buf->data + buf->read_pos * buf->element_size;
*data_ptr_bytes_1 = margin * buf->element_size;
*data_ptr_2 = buf->data;
*data_ptr_bytes_2 = (read_elements - margin) * buf->element_size;
} else {
*data_ptr_1 = buf->data + buf->read_pos * buf->element_size;
*data_ptr_bytes_1 = read_elements * buf->element_size;
*data_ptr_2 = NULL;
*data_ptr_bytes_2 = 0;
}
return read_elements;
}
size_t WebRtc_available_write(const RingBuffer* self) {
if (!self) {
return 0;
}
return self->element_count - WebRtc_available_read(self);
}
void WebRtcAec_ProcessFrame(aec_t *aec,
const short *nearend,
const short *nearendH,
int knownDelay)
{
// For each frame the process is as follows:
// 1) If the system_delay indicates on being too small for processing a
// frame we stuff the buffer with enough data for 10 ms.
// 2) Adjust the buffer to the system delay, by moving the read pointer.
// 3) If we can't move read pointer due to buffer size limitations we
// flush/stuff the buffer.
// 4) Process as many partitions as possible.
// 5) Update the |system_delay| with respect to a full frame of FRAME_LEN
// samples. Even though we will have data left to process (we work with
// partitions) we consider updating a whole frame, since that's the
// amount of data we input and output in audio_processing.
// TODO(bjornv): Investigate how we should round the delay difference; right
// now we know that incoming |knownDelay| is underestimated when it's less
// than |aec->knownDelay|. We therefore, round (-32) in that direction. In
// the other direction, we don't have this situation, but might flush one
// partition too little. This can cause non-causality, which should be
// investigated. Maybe, allow for a non-symmetric rounding, like -16.
int move_elements = (aec->knownDelay - knownDelay - 32) / PART_LEN;
int moved_elements = 0;
// TODO(bjornv): Change the near-end buffer handling to be the same as for
// far-end, that is, with a near_pre_buf.
// Buffer the near-end frame.
WebRtc_WriteBuffer(aec->nearFrBuf, nearend, FRAME_LEN);
// For H band
if (aec->sampFreq == 32000) {
WebRtc_WriteBuffer(aec->nearFrBufH, nearendH, FRAME_LEN);
}
// 1) At most we process |aec->mult|+1 partitions in 10 ms. Make sure we
// have enough far-end data for that by stuffing the buffer if the
// |system_delay| indicates others.
if (aec->system_delay < FRAME_LEN) {
// We don't have enough data so we rewind 10 ms.
WebRtcAec_MoveFarReadPtr(aec, -(aec->mult + 1));
}
// 2) Compensate for a possible change in the system delay.
WebRtc_MoveReadPtr(aec->far_buf_windowed, move_elements);
moved_elements = WebRtc_MoveReadPtr(aec->far_buf, move_elements);
aec->knownDelay -= moved_elements * PART_LEN;
#ifdef WEBRTC_AEC_DEBUG_DUMP
WebRtc_MoveReadPtr(aec->far_time_buf, move_elements);
#endif
// 4) Process as many blocks as possible.
while (WebRtc_available_read(aec->nearFrBuf) >= PART_LEN) {
ProcessBlock(aec);
}
// 5) Update system delay with respect to the entire frame.
aec->system_delay -= FRAME_LEN;
}
// only buffer L band for farend
int32_t WebRtcAec_BufferFarend(void* aecInst,
const float* farend,
size_t nrOfSamples) {
Aec* aecpc = aecInst;
size_t newNrOfSamples = nrOfSamples;
float new_farend[MAX_RESAMP_LEN];
const float* farend_ptr = farend;
// Get any error caused by buffering the farend signal.
int32_t error_code = WebRtcAec_GetBufferFarendError(aecInst, farend,
nrOfSamples);
if (error_code != 0)
return error_code;
if (aecpc->skewMode == kAecTrue && aecpc->resample == kAecTrue) {
// Resample and get a new number of samples
WebRtcAec_ResampleLinear(aecpc->resampler,
farend,
nrOfSamples,
aecpc->skew,
new_farend,
&newNrOfSamples);
farend_ptr = new_farend;
}
aecpc->farend_started = 1;
WebRtcAec_SetSystemDelay(
aecpc->aec, WebRtcAec_system_delay(aecpc->aec) + (int)newNrOfSamples);
// Write the time-domain data to |far_pre_buf|.
WebRtc_WriteBuffer(aecpc->far_pre_buf, farend_ptr, newNrOfSamples);
// Transform to frequency domain if we have enough data.
while (WebRtc_available_read(aecpc->far_pre_buf) >= PART_LEN2) {
// We have enough data to pass to the FFT, hence read PART_LEN2 samples.
{
float* ptmp = NULL;
float tmp[PART_LEN2];
WebRtc_ReadBuffer(aecpc->far_pre_buf, (void**)&ptmp, tmp, PART_LEN2);
WebRtcAec_BufferFarendPartition(aecpc->aec, ptmp);
#ifdef WEBRTC_AEC_DEBUG_DUMP
WebRtc_WriteBuffer(
WebRtcAec_far_time_buf(aecpc->aec), &ptmp[PART_LEN], 1);
#endif
}
// Rewind |far_pre_buf| PART_LEN samples for overlap before continuing.
WebRtc_MoveReadPtr(aecpc->far_pre_buf, -PART_LEN);
}
return 0;
}
int WebRtc_MoveReadPtr(RingBuffer* self, int element_count) {
if (!self) {
return 0;
}
{
// We need to be able to take care of negative changes, hence use "int"
// instead of "size_t".
const int free_elements = (int) WebRtc_available_write(self);
const int readable_elements = (int) WebRtc_available_read(self);
int read_pos = (int) self->read_pos;
if (element_count > readable_elements) {
element_count = readable_elements;
}
if (element_count < -free_elements) {
element_count = -free_elements;
}
read_pos += element_count;
if (read_pos > (int) self->element_count) {
// Buffer wrap around. Restart read position and wrap indicator.
read_pos -= (int) self->element_count;
self->rw_wrap = SAME_WRAP;
}
if (read_pos < 0) {
// Buffer wrap around. Restart read position and wrap indicator.
read_pos += (int) self->element_count;
self->rw_wrap = DIFF_WRAP;
}
self->read_pos = (size_t) read_pos;
return element_count;
}
}
// only buffer L band for farend
int32_t WebRtcAec_BufferFarend(void *aecInst, const int16_t *farend,
int16_t nrOfSamples)
{
aecpc_t *aecpc = aecInst;
int32_t retVal = 0;
int newNrOfSamples = (int) nrOfSamples;
short newFarend[MAX_RESAMP_LEN];
const int16_t* farend_ptr = farend;
float tmp_farend[MAX_RESAMP_LEN];
const float* farend_float = tmp_farend;
float skew;
int i = 0;
if (aecpc == NULL) {
return -1;
}
if (farend == NULL) {
aecpc->lastError = AEC_NULL_POINTER_ERROR;
return -1;
}
if (aecpc->initFlag != initCheck) {
aecpc->lastError = AEC_UNINITIALIZED_ERROR;
return -1;
}
// number of samples == 160 for SWB input
if (nrOfSamples != 80 && nrOfSamples != 160) {
aecpc->lastError = AEC_BAD_PARAMETER_ERROR;
return -1;
}
skew = aecpc->skew;
if (aecpc->skewMode == kAecTrue && aecpc->resample == kAecTrue) {
// Resample and get a new number of samples
WebRtcAec_ResampleLinear(aecpc->resampler, farend, nrOfSamples, skew,
newFarend, &newNrOfSamples);
farend_ptr = (const int16_t*) newFarend;
}
WebRtcAec_SetSystemDelay(aecpc->aec, WebRtcAec_system_delay(aecpc->aec) +
newNrOfSamples);
#ifdef WEBRTC_AEC_DEBUG_DUMP
WebRtc_WriteBuffer(aecpc->far_pre_buf_s16, farend_ptr,
(size_t) newNrOfSamples);
#endif
// Cast to float and write the time-domain data to |far_pre_buf|.
for (i = 0; i < newNrOfSamples; i++) {
tmp_farend[i] = (float) farend_ptr[i];
}
WebRtc_WriteBuffer(aecpc->far_pre_buf, farend_float,
(size_t) newNrOfSamples);
// Transform to frequency domain if we have enough data.
while (WebRtc_available_read(aecpc->far_pre_buf) >= PART_LEN2) {
// We have enough data to pass to the FFT, hence read PART_LEN2 samples.
WebRtc_ReadBuffer(aecpc->far_pre_buf, (void**) &farend_float, tmp_farend,
PART_LEN2);
WebRtcAec_BufferFarendPartition(aecpc->aec, farend_float);
// Rewind |far_pre_buf| PART_LEN samples for overlap before continuing.
WebRtc_MoveReadPtr(aecpc->far_pre_buf, -PART_LEN);
#ifdef WEBRTC_AEC_DEBUG_DUMP
WebRtc_ReadBuffer(aecpc->far_pre_buf_s16, (void**) &farend_ptr, newFarend,
PART_LEN2);
WebRtc_WriteBuffer(WebRtcAec_far_time_buf(aecpc->aec),
&farend_ptr[PART_LEN], 1);
WebRtc_MoveReadPtr(aecpc->far_pre_buf_s16, -PART_LEN);
#endif
}
return retVal;
}
WebRtc_Word32 WebRtcAec_Process(void *aecInst, const WebRtc_Word16 *nearend,
const WebRtc_Word16 *nearendH, WebRtc_Word16 *out, WebRtc_Word16 *outH,
WebRtc_Word16 nrOfSamples, WebRtc_Word16 msInSndCardBuf, WebRtc_Word32 skew)
{
aecpc_t *aecpc = aecInst;
WebRtc_Word32 retVal = 0;
short i;
short nBlocks10ms;
short nFrames;
// Limit resampling to doubling/halving of signal
const float minSkewEst = -0.5f;
const float maxSkewEst = 1.0f;
if (aecpc == NULL) {
return -1;
}
if (nearend == NULL) {
aecpc->lastError = AEC_NULL_POINTER_ERROR;
return -1;
}
if (out == NULL) {
aecpc->lastError = AEC_NULL_POINTER_ERROR;
return -1;
}
if (aecpc->initFlag != initCheck) {
aecpc->lastError = AEC_UNINITIALIZED_ERROR;
return -1;
}
// number of samples == 160 for SWB input
if (nrOfSamples != 80 && nrOfSamples != 160) {
aecpc->lastError = AEC_BAD_PARAMETER_ERROR;
return -1;
}
// Check for valid pointers based on sampling rate
if (aecpc->sampFreq == 32000 && nearendH == NULL) {
aecpc->lastError = AEC_NULL_POINTER_ERROR;
return -1;
}
if (msInSndCardBuf < 0) {
msInSndCardBuf = 0;
aecpc->lastError = AEC_BAD_PARAMETER_WARNING;
retVal = -1;
}
else if (msInSndCardBuf > 500) {
msInSndCardBuf = 500;
aecpc->lastError = AEC_BAD_PARAMETER_WARNING;
retVal = -1;
}
// TODO(andrew): we need to investigate if this +10 is really wanted.
msInSndCardBuf += 10;
aecpc->msInSndCardBuf = msInSndCardBuf;
if (aecpc->skewMode == kAecTrue) {
if (aecpc->skewFrCtr < 25) {
aecpc->skewFrCtr++;
}
else {
retVal = WebRtcAec_GetSkew(aecpc->resampler, skew, &aecpc->skew);
if (retVal == -1) {
aecpc->skew = 0;
aecpc->lastError = AEC_BAD_PARAMETER_WARNING;
}
aecpc->skew /= aecpc->sampFactor*nrOfSamples;
if (aecpc->skew < 1.0e-3 && aecpc->skew > -1.0e-3) {
aecpc->resample = kAecFalse;
}
else {
aecpc->resample = kAecTrue;
}
if (aecpc->skew < minSkewEst) {
aecpc->skew = minSkewEst;
}
else if (aecpc->skew > maxSkewEst) {
aecpc->skew = maxSkewEst;
}
#ifdef WEBRTC_AEC_DEBUG_DUMP
fwrite(&aecpc->skew, sizeof(aecpc->skew), 1, aecpc->skewFile);
#endif
}
}
nFrames = nrOfSamples / FRAME_LEN;
nBlocks10ms = nFrames / aecpc->aec->mult;
if (aecpc->ECstartup) {
if (nearend != out) {
// Only needed if they don't already point to the same place.
memcpy(out, nearend, sizeof(short) * nrOfSamples);
}
// The AEC is in the start up mode
// AEC is disabled until the system delay is OK
// Mechanism to ensure that the system delay is reasonably stable.
if (aecpc->checkBuffSize) {
aecpc->checkBufSizeCtr++;
// Before we fill up the far-end buffer we require the system delay
// to be stable (+/-8 ms) compared to the first value. This
// comparison is made during the following 6 consecutive 10 ms
// blocks. If it seems to be stable then we start to fill up the
// far-end buffer.
if (aecpc->counter == 0) {
aecpc->firstVal = aecpc->msInSndCardBuf;
aecpc->sum = 0;
}
if (abs(aecpc->firstVal - aecpc->msInSndCardBuf) <
WEBRTC_SPL_MAX(0.2 * aecpc->msInSndCardBuf, sampMsNb)) {
aecpc->sum += aecpc->msInSndCardBuf;
aecpc->counter++;
}
else {
aecpc->counter = 0;
}
if (aecpc->counter * nBlocks10ms >= 6) {
// The far-end buffer size is determined in partitions of
// PART_LEN samples. Use 75% of the average value of the system
// delay as buffer size to start with.
aecpc->bufSizeStart = WEBRTC_SPL_MIN((3 * aecpc->sum *
aecpc->aec->mult * 8) / (4 * aecpc->counter * PART_LEN),
kMaxBufSizeStart);
// Buffer size has now been determined.
aecpc->checkBuffSize = 0;
}
if (aecpc->checkBufSizeCtr * nBlocks10ms > 50) {
// For really bad systems, don't disable the echo canceller for
// more than 0.5 sec.
aecpc->bufSizeStart = WEBRTC_SPL_MIN((aecpc->msInSndCardBuf *
aecpc->aec->mult * 3) / 40, kMaxBufSizeStart);
aecpc->checkBuffSize = 0;
}
}
// If |checkBuffSize| changed in the if-statement above.
if (!aecpc->checkBuffSize) {
// The system delay is now reasonably stable (or has been unstable
// for too long). When the far-end buffer is filled with
// approximately the same amount of data as reported by the system
// we end the startup phase.
int overhead_elements = aecpc->aec->system_delay / PART_LEN -
aecpc->bufSizeStart;
if (overhead_elements == 0) {
// Enable the AEC
aecpc->ECstartup = 0;
} else if (overhead_elements > 0) {
// TODO(bjornv): Do we need a check on how much we actually
// moved the read pointer? It should always be possible to move
// the pointer |overhead_elements| since we have only added data
// to the buffer and no delay compensation nor AEC processing
// has been done.
WebRtcAec_MoveFarReadPtr(aecpc->aec, overhead_elements);
// Enable the AEC
aecpc->ECstartup = 0;
}
}
} else {
// AEC is enabled.
int out_elements = 0;
EstBufDelay(aecpc);
// Note that 1 frame is supported for NB and 2 frames for WB.
for (i = 0; i < nFrames; i++) {
int16_t* out_ptr = NULL;
int16_t out_tmp[FRAME_LEN];
// Call the AEC.
WebRtcAec_ProcessFrame(aecpc->aec,
&nearend[FRAME_LEN * i],
&nearendH[FRAME_LEN * i],
aecpc->knownDelay);
// TODO(bjornv): Re-structure such that we don't have to pass
// |aecpc->knownDelay| as input. Change name to something like
// |system_buffer_diff|.
// Stuff the out buffer if we have less than a frame to output.
// This should only happen for the first frame.
out_elements = (int) WebRtc_available_read(aecpc->aec->outFrBuf);
if (out_elements < FRAME_LEN) {
WebRtc_MoveReadPtr(aecpc->aec->outFrBuf,
out_elements - FRAME_LEN);
if (aecpc->sampFreq == 32000) {
WebRtc_MoveReadPtr(aecpc->aec->outFrBufH,
out_elements - FRAME_LEN);
}
}
// Obtain an output frame.
WebRtc_ReadBuffer(aecpc->aec->outFrBuf, (void**) &out_ptr,
out_tmp, FRAME_LEN);
memcpy(&out[FRAME_LEN * i], out_ptr, sizeof(int16_t) * FRAME_LEN);
// For H band
if (aecpc->sampFreq == 32000) {
WebRtc_ReadBuffer(aecpc->aec->outFrBufH, (void**) &out_ptr,
out_tmp, FRAME_LEN);
memcpy(&outH[FRAME_LEN * i], out_ptr,
sizeof(int16_t) * FRAME_LEN);
}
}
}
#ifdef WEBRTC_AEC_DEBUG_DUMP
{
int16_t far_buf_size_ms = (int16_t) (aecpc->aec->system_delay /
(sampMsNb * aecpc->aec->mult));
fwrite(&far_buf_size_ms, 2, 1, aecpc->bufFile);
fwrite(&(aecpc->knownDelay), sizeof(aecpc->knownDelay), 1, aecpc->delayFile);
}
#endif
return retVal;
}
// only buffer L band for farend
int32_t WebRtcAec_BufferFarend(void* aecInst,
const float* farend,
int16_t nrOfSamples) {
Aec* aecpc = aecInst;
int newNrOfSamples = (int)nrOfSamples;
float new_farend[MAX_RESAMP_LEN];
const float* farend_ptr = farend;
if (farend == NULL) {
aecpc->lastError = AEC_NULL_POINTER_ERROR;
return -1;
}
if (aecpc->initFlag != initCheck) {
aecpc->lastError = AEC_UNINITIALIZED_ERROR;
return -1;
}
// number of samples == 160 for SWB input
if (nrOfSamples != 80 && nrOfSamples != 160) {
aecpc->lastError = AEC_BAD_PARAMETER_ERROR;
return -1;
}
if (aecpc->skewMode == kAecTrue && aecpc->resample == kAecTrue) {
// Resample and get a new number of samples
WebRtcAec_ResampleLinear(aecpc->resampler,
farend,
nrOfSamples,
aecpc->skew,
new_farend,
&newNrOfSamples);
farend_ptr = new_farend;
}
aecpc->farend_started = 1;
WebRtcAec_SetSystemDelay(aecpc->aec,
WebRtcAec_system_delay(aecpc->aec) + newNrOfSamples);
// Write the time-domain data to |far_pre_buf|.
WebRtc_WriteBuffer(aecpc->far_pre_buf, farend_ptr, (size_t)newNrOfSamples);
// Transform to frequency domain if we have enough data.
while (WebRtc_available_read(aecpc->far_pre_buf) >= PART_LEN2) {
// We have enough data to pass to the FFT, hence read PART_LEN2 samples.
{
float* ptmp = NULL;
float tmp[PART_LEN2];
WebRtc_ReadBuffer(aecpc->far_pre_buf, (void**)&ptmp, tmp, PART_LEN2);
WebRtcAec_BufferFarendPartition(aecpc->aec, ptmp);
#ifdef WEBRTC_AEC_DEBUG_DUMP
WebRtc_WriteBuffer(
WebRtcAec_far_time_buf(aecpc->aec), &ptmp[PART_LEN], 1);
#endif
}
// Rewind |far_pre_buf| PART_LEN samples for overlap before continuing.
WebRtc_MoveReadPtr(aecpc->far_pre_buf, -PART_LEN);
}
return 0;
}
static void NonLinearProcessing(aec_t *aec, short *output, short *outputH)
{
float efw[2][PART_LEN1], dfw[2][PART_LEN1], xfw[2][PART_LEN1];
complex_t comfortNoiseHband[PART_LEN1];
float fft[PART_LEN2];
float scale, dtmp;
float nlpGainHband;
int i, j, pos;
// Coherence and non-linear filter
float cohde[PART_LEN1], cohxd[PART_LEN1];
float hNlDeAvg, hNlXdAvg;
float hNl[PART_LEN1];
float hNlPref[PREF_BAND_SIZE];
float hNlFb = 0, hNlFbLow = 0;
const float prefBandQuant = 0.75f, prefBandQuantLow = 0.5f;
const int prefBandSize = PREF_BAND_SIZE / aec->mult;
const int minPrefBand = 4 / aec->mult;
// Near and error power sums
float sdSum = 0, seSum = 0;
// Power estimate smoothing coefficients
const float gCoh[2][2] = {{0.9f, 0.1f}, {0.93f, 0.07f}};
const float *ptrGCoh = gCoh[aec->mult - 1];
// Filter energy
float wfEnMax = 0, wfEn = 0;
const int delayEstInterval = 10 * aec->mult;
float* xfw_ptr = NULL;
aec->delayEstCtr++;
if (aec->delayEstCtr == delayEstInterval) {
aec->delayEstCtr = 0;
}
// initialize comfort noise for H band
memset(comfortNoiseHband, 0, sizeof(comfortNoiseHband));
nlpGainHband = (float)0.0;
dtmp = (float)0.0;
// Measure energy in each filter partition to determine delay.
// TODO: Spread by computing one partition per block?
if (aec->delayEstCtr == 0) {
wfEnMax = 0;
aec->delayIdx = 0;
for (i = 0; i < NR_PART; i++) {
pos = i * PART_LEN1;
wfEn = 0;
for (j = 0; j < PART_LEN1; j++) {
wfEn += aec->wfBuf[0][pos + j] * aec->wfBuf[0][pos + j] +
aec->wfBuf[1][pos + j] * aec->wfBuf[1][pos + j];
}
if (wfEn > wfEnMax) {
wfEnMax = wfEn;
aec->delayIdx = i;
}
}
}
// We should always have at least one element stored in |far_buf|.
assert(WebRtc_available_read(aec->far_buf_windowed) > 0);
// NLP
WebRtc_ReadBuffer(aec->far_buf_windowed, (void**) &xfw_ptr, &xfw[0][0], 1);
// TODO(bjornv): Investigate if we can reuse |far_buf_windowed| instead of
// |xfwBuf|.
// Buffer far.
memcpy(aec->xfwBuf, xfw_ptr, sizeof(float) * 2 * PART_LEN1);
// Use delayed far.
memcpy(xfw, aec->xfwBuf + aec->delayIdx * PART_LEN1, sizeof(xfw));
// Windowed near fft
for (i = 0; i < PART_LEN; i++) {
fft[i] = aec->dBuf[i] * sqrtHanning[i];
fft[PART_LEN + i] = aec->dBuf[PART_LEN + i] * sqrtHanning[PART_LEN - i];
}
aec_rdft_forward_128(fft);
dfw[1][0] = 0;
dfw[1][PART_LEN] = 0;
dfw[0][0] = fft[0];
dfw[0][PART_LEN] = fft[1];
for (i = 1; i < PART_LEN; i++) {
dfw[0][i] = fft[2 * i];
dfw[1][i] = fft[2 * i + 1];
}
// Windowed error fft
for (i = 0; i < PART_LEN; i++) {
fft[i] = aec->eBuf[i] * sqrtHanning[i];
fft[PART_LEN + i] = aec->eBuf[PART_LEN + i] * sqrtHanning[PART_LEN - i];
}
aec_rdft_forward_128(fft);
efw[1][0] = 0;
efw[1][PART_LEN] = 0;
efw[0][0] = fft[0];
efw[0][PART_LEN] = fft[1];
for (i = 1; i < PART_LEN; i++) {
efw[0][i] = fft[2 * i];
efw[1][i] = fft[2 * i + 1];
}
// Smoothed PSD
for (i = 0; i < PART_LEN1; i++) {
aec->sd[i] = ptrGCoh[0] * aec->sd[i] + ptrGCoh[1] *
(dfw[0][i] * dfw[0][i] + dfw[1][i] * dfw[1][i]);
aec->se[i] = ptrGCoh[0] * aec->se[i] + ptrGCoh[1] *
(efw[0][i] * efw[0][i] + efw[1][i] * efw[1][i]);
// We threshold here to protect against the ill-effects of a zero farend.
// The threshold is not arbitrarily chosen, but balances protection and
// adverse interaction with the algorithm's tuning.
// TODO: investigate further why this is so sensitive.
aec->sx[i] = ptrGCoh[0] * aec->sx[i] + ptrGCoh[1] *
WEBRTC_SPL_MAX(xfw[0][i] * xfw[0][i] + xfw[1][i] * xfw[1][i], 15);
aec->sde[i][0] = ptrGCoh[0] * aec->sde[i][0] + ptrGCoh[1] *
(dfw[0][i] * efw[0][i] + dfw[1][i] * efw[1][i]);
aec->sde[i][1] = ptrGCoh[0] * aec->sde[i][1] + ptrGCoh[1] *
(dfw[0][i] * efw[1][i] - dfw[1][i] * efw[0][i]);
aec->sxd[i][0] = ptrGCoh[0] * aec->sxd[i][0] + ptrGCoh[1] *
(dfw[0][i] * xfw[0][i] + dfw[1][i] * xfw[1][i]);
aec->sxd[i][1] = ptrGCoh[0] * aec->sxd[i][1] + ptrGCoh[1] *
(dfw[0][i] * xfw[1][i] - dfw[1][i] * xfw[0][i]);
sdSum += aec->sd[i];
seSum += aec->se[i];
}
// Divergent filter safeguard.
if (aec->divergeState == 0) {
if (seSum > sdSum) {
aec->divergeState = 1;
}
}
else {
if (seSum * 1.05f < sdSum) {
aec->divergeState = 0;
}
}
if (aec->divergeState == 1) {
memcpy(efw, dfw, sizeof(efw));
}
// Reset if error is significantly larger than nearend (13 dB).
if (seSum > (19.95f * sdSum)) {
memset(aec->wfBuf, 0, sizeof(aec->wfBuf));
}
// Subband coherence
for (i = 0; i < PART_LEN1; i++) {
cohde[i] = (aec->sde[i][0] * aec->sde[i][0] + aec->sde[i][1] * aec->sde[i][1]) /
(aec->sd[i] * aec->se[i] + 1e-10f);
cohxd[i] = (aec->sxd[i][0] * aec->sxd[i][0] + aec->sxd[i][1] * aec->sxd[i][1]) /
(aec->sx[i] * aec->sd[i] + 1e-10f);
}
hNlXdAvg = 0;
for (i = minPrefBand; i < prefBandSize + minPrefBand; i++) {
hNlXdAvg += cohxd[i];
}
hNlXdAvg /= prefBandSize;
hNlXdAvg = 1 - hNlXdAvg;
hNlDeAvg = 0;
for (i = minPrefBand; i < prefBandSize + minPrefBand; i++) {
hNlDeAvg += cohde[i];
}
hNlDeAvg /= prefBandSize;
if (hNlXdAvg < 0.75f && hNlXdAvg < aec->hNlXdAvgMin) {
aec->hNlXdAvgMin = hNlXdAvg;
}
if (hNlDeAvg > 0.98f && hNlXdAvg > 0.9f) {
aec->stNearState = 1;
}
else if (hNlDeAvg < 0.95f || hNlXdAvg < 0.8f) {
aec->stNearState = 0;
}
if (aec->hNlXdAvgMin == 1) {
aec->echoState = 0;
aec->overDrive = aec->minOverDrive;
if (aec->stNearState == 1) {
memcpy(hNl, cohde, sizeof(hNl));
hNlFb = hNlDeAvg;
hNlFbLow = hNlDeAvg;
}
else {
for (i = 0; i < PART_LEN1; i++) {
hNl[i] = 1 - cohxd[i];
}
hNlFb = hNlXdAvg;
hNlFbLow = hNlXdAvg;
}
}
else {
if (aec->stNearState == 1) {
aec->echoState = 0;
memcpy(hNl, cohde, sizeof(hNl));
hNlFb = hNlDeAvg;
hNlFbLow = hNlDeAvg;
}
else {
aec->echoState = 1;
for (i = 0; i < PART_LEN1; i++) {
hNl[i] = WEBRTC_SPL_MIN(cohde[i], 1 - cohxd[i]);
}
// Select an order statistic from the preferred bands.
// TODO: Using quicksort now, but a selection algorithm may be preferred.
memcpy(hNlPref, &hNl[minPrefBand], sizeof(float) * prefBandSize);
qsort(hNlPref, prefBandSize, sizeof(float), CmpFloat);
hNlFb = hNlPref[(int)floor(prefBandQuant * (prefBandSize - 1))];
hNlFbLow = hNlPref[(int)floor(prefBandQuantLow * (prefBandSize - 1))];
}
}
// Track the local filter minimum to determine suppression overdrive.
if (hNlFbLow < 0.6f && hNlFbLow < aec->hNlFbLocalMin) {
aec->hNlFbLocalMin = hNlFbLow;
aec->hNlFbMin = hNlFbLow;
aec->hNlNewMin = 1;
aec->hNlMinCtr = 0;
}
aec->hNlFbLocalMin = WEBRTC_SPL_MIN(aec->hNlFbLocalMin + 0.0008f / aec->mult, 1);
aec->hNlXdAvgMin = WEBRTC_SPL_MIN(aec->hNlXdAvgMin + 0.0006f / aec->mult, 1);
if (aec->hNlNewMin == 1) {
aec->hNlMinCtr++;
}
if (aec->hNlMinCtr == 2) {
aec->hNlNewMin = 0;
aec->hNlMinCtr = 0;
aec->overDrive = WEBRTC_SPL_MAX(aec->targetSupp /
((float)log(aec->hNlFbMin + 1e-10f) + 1e-10f), aec->minOverDrive);
}
// Smooth the overdrive.
if (aec->overDrive < aec->overDriveSm) {
aec->overDriveSm = 0.99f * aec->overDriveSm + 0.01f * aec->overDrive;
}
else {
aec->overDriveSm = 0.9f * aec->overDriveSm + 0.1f * aec->overDrive;
}
WebRtcAec_OverdriveAndSuppress(aec, hNl, hNlFb, efw);
// Add comfort noise.
ComfortNoise(aec, efw, comfortNoiseHband, aec->noisePow, hNl);
// TODO(bjornv): Investigate how to take the windowing below into account if
// needed.
if (aec->metricsMode == 1) {
// Note that we have a scaling by two in the time domain |eBuf|.
// In addition the time domain signal is windowed before transformation,
// losing half the energy on the average. We take care of the first
// scaling only in UpdateMetrics().
UpdateLevel(&aec->nlpoutlevel, efw);
}
// Inverse error fft.
fft[0] = efw[0][0];
fft[1] = efw[0][PART_LEN];
for (i = 1; i < PART_LEN; i++) {
fft[2*i] = efw[0][i];
// Sign change required by Ooura fft.
fft[2*i + 1] = -efw[1][i];
}
aec_rdft_inverse_128(fft);
// Overlap and add to obtain output.
scale = 2.0f / PART_LEN2;
for (i = 0; i < PART_LEN; i++) {
fft[i] *= scale; // fft scaling
fft[i] = fft[i]*sqrtHanning[i] + aec->outBuf[i];
// Saturation protection
output[i] = (short)WEBRTC_SPL_SAT(WEBRTC_SPL_WORD16_MAX, fft[i],
WEBRTC_SPL_WORD16_MIN);
fft[PART_LEN + i] *= scale; // fft scaling
aec->outBuf[i] = fft[PART_LEN + i] * sqrtHanning[PART_LEN - i];
}
// For H band
if (aec->sampFreq == 32000) {
// H band gain
// average nlp over low band: average over second half of freq spectrum
// (4->8khz)
GetHighbandGain(hNl, &nlpGainHband);
// Inverse comfort_noise
if (flagHbandCn == 1) {
fft[0] = comfortNoiseHband[0][0];
fft[1] = comfortNoiseHband[PART_LEN][0];
for (i = 1; i < PART_LEN; i++) {
fft[2*i] = comfortNoiseHband[i][0];
fft[2*i + 1] = comfortNoiseHband[i][1];
}
aec_rdft_inverse_128(fft);
scale = 2.0f / PART_LEN2;
}
// compute gain factor
for (i = 0; i < PART_LEN; i++) {
dtmp = (float)aec->dBufH[i];
dtmp = (float)dtmp * nlpGainHband; // for variable gain
// add some comfort noise where Hband is attenuated
if (flagHbandCn == 1) {
fft[i] *= scale; // fft scaling
dtmp += cnScaleHband * fft[i];
}
// Saturation protection
outputH[i] = (short)WEBRTC_SPL_SAT(WEBRTC_SPL_WORD16_MAX, dtmp,
WEBRTC_SPL_WORD16_MIN);
}
}
// Copy the current block to the old position.
memcpy(aec->dBuf, aec->dBuf + PART_LEN, sizeof(float) * PART_LEN);
memcpy(aec->eBuf, aec->eBuf + PART_LEN, sizeof(float) * PART_LEN);
// Copy the current block to the old position for H band
if (aec->sampFreq == 32000) {
memcpy(aec->dBufH, aec->dBufH + PART_LEN, sizeof(float) * PART_LEN);
}
memmove(aec->xfwBuf + PART_LEN1, aec->xfwBuf, sizeof(aec->xfwBuf) -
sizeof(complex_t) * PART_LEN1);
}
static void ProcessBlock(aec_t* aec) {
int i;
float d[PART_LEN], y[PART_LEN], e[PART_LEN], dH[PART_LEN];
float scale;
float fft[PART_LEN2];
float xf[2][PART_LEN1], yf[2][PART_LEN1], ef[2][PART_LEN1];
float df[2][PART_LEN1];
float far_spectrum = 0.0f;
float near_spectrum = 0.0f;
float abs_far_spectrum[PART_LEN1];
float abs_near_spectrum[PART_LEN1];
const float gPow[2] = {0.9f, 0.1f};
// Noise estimate constants.
const int noiseInitBlocks = 500 * aec->mult;
const float step = 0.1f;
const float ramp = 1.0002f;
const float gInitNoise[2] = {0.999f, 0.001f};
int16_t nearend[PART_LEN];
int16_t* nearend_ptr = NULL;
int16_t output[PART_LEN];
int16_t outputH[PART_LEN];
float* xf_ptr = NULL;
memset(dH, 0, sizeof(dH));
if (aec->sampFreq == 32000) {
// Get the upper band first so we can reuse |nearend|.
WebRtc_ReadBuffer(aec->nearFrBufH,
(void**) &nearend_ptr,
nearend,
PART_LEN);
for (i = 0; i < PART_LEN; i++) {
dH[i] = (float) (nearend_ptr[i]);
}
memcpy(aec->dBufH + PART_LEN, dH, sizeof(float) * PART_LEN);
}
WebRtc_ReadBuffer(aec->nearFrBuf, (void**) &nearend_ptr, nearend, PART_LEN);
// ---------- Ooura fft ----------
// Concatenate old and new nearend blocks.
for (i = 0; i < PART_LEN; i++) {
d[i] = (float) (nearend_ptr[i]);
}
memcpy(aec->dBuf + PART_LEN, d, sizeof(float) * PART_LEN);
#ifdef WEBRTC_AEC_DEBUG_DUMP
{
int16_t farend[PART_LEN];
int16_t* farend_ptr = NULL;
WebRtc_ReadBuffer(aec->far_time_buf, (void**) &farend_ptr, farend, 1);
(void)fwrite(farend_ptr, sizeof(int16_t), PART_LEN, aec->farFile);
(void)fwrite(nearend_ptr, sizeof(int16_t), PART_LEN, aec->nearFile);
}
#endif
// We should always have at least one element stored in |far_buf|.
assert(WebRtc_available_read(aec->far_buf) > 0);
WebRtc_ReadBuffer(aec->far_buf, (void**) &xf_ptr, &xf[0][0], 1);
// Near fft
memcpy(fft, aec->dBuf, sizeof(float) * PART_LEN2);
TimeToFrequency(fft, df, 0);
// Power smoothing
for (i = 0; i < PART_LEN1; i++) {
far_spectrum = (xf_ptr[i] * xf_ptr[i]) +
(xf_ptr[PART_LEN1 + i] * xf_ptr[PART_LEN1 + i]);
aec->xPow[i] = gPow[0] * aec->xPow[i] + gPow[1] * NR_PART * far_spectrum;
// Calculate absolute spectra
abs_far_spectrum[i] = sqrtf(far_spectrum);
near_spectrum = df[0][i] * df[0][i] + df[1][i] * df[1][i];
aec->dPow[i] = gPow[0] * aec->dPow[i] + gPow[1] * near_spectrum;
// Calculate absolute spectra
abs_near_spectrum[i] = sqrtf(near_spectrum);
}
// Estimate noise power. Wait until dPow is more stable.
if (aec->noiseEstCtr > 50) {
for (i = 0; i < PART_LEN1; i++) {
if (aec->dPow[i] < aec->dMinPow[i]) {
aec->dMinPow[i] = (aec->dPow[i] + step * (aec->dMinPow[i] -
aec->dPow[i])) * ramp;
}
else {
aec->dMinPow[i] *= ramp;
}
}
}
// Smooth increasing noise power from zero at the start,
// to avoid a sudden burst of comfort noise.
if (aec->noiseEstCtr < noiseInitBlocks) {
aec->noiseEstCtr++;
for (i = 0; i < PART_LEN1; i++) {
if (aec->dMinPow[i] > aec->dInitMinPow[i]) {
aec->dInitMinPow[i] = gInitNoise[0] * aec->dInitMinPow[i] +
gInitNoise[1] * aec->dMinPow[i];
}
else {
aec->dInitMinPow[i] = aec->dMinPow[i];
}
}
aec->noisePow = aec->dInitMinPow;
}
else {
aec->noisePow = aec->dMinPow;
}
// Block wise delay estimation used for logging
if (aec->delay_logging_enabled) {
int delay_estimate = 0;
// Estimate the delay
delay_estimate = WebRtc_DelayEstimatorProcessFloat(aec->delay_estimator,
abs_far_spectrum,
abs_near_spectrum,
PART_LEN1);
if (delay_estimate >= 0) {
// Update delay estimate buffer.
aec->delay_histogram[delay_estimate]++;
}
}
// Update the xfBuf block position.
aec->xfBufBlockPos--;
if (aec->xfBufBlockPos == -1) {
aec->xfBufBlockPos = NR_PART - 1;
}
// Buffer xf
memcpy(aec->xfBuf[0] + aec->xfBufBlockPos * PART_LEN1, xf_ptr,
sizeof(float) * PART_LEN1);
memcpy(aec->xfBuf[1] + aec->xfBufBlockPos * PART_LEN1, &xf_ptr[PART_LEN1],
sizeof(float) * PART_LEN1);
memset(yf, 0, sizeof(yf));
// Filter far
WebRtcAec_FilterFar(aec, yf);
// Inverse fft to obtain echo estimate and error.
fft[0] = yf[0][0];
fft[1] = yf[0][PART_LEN];
for (i = 1; i < PART_LEN; i++) {
fft[2 * i] = yf[0][i];
fft[2 * i + 1] = yf[1][i];
}
aec_rdft_inverse_128(fft);
scale = 2.0f / PART_LEN2;
for (i = 0; i < PART_LEN; i++) {
y[i] = fft[PART_LEN + i] * scale; // fft scaling
}
for (i = 0; i < PART_LEN; i++) {
e[i] = d[i] - y[i];
}
// Error fft
memcpy(aec->eBuf + PART_LEN, e, sizeof(float) * PART_LEN);
memset(fft, 0, sizeof(float) * PART_LEN);
memcpy(fft + PART_LEN, e, sizeof(float) * PART_LEN);
// TODO(bjornv): Change to use TimeToFrequency().
aec_rdft_forward_128(fft);
ef[1][0] = 0;
ef[1][PART_LEN] = 0;
ef[0][0] = fft[0];
ef[0][PART_LEN] = fft[1];
for (i = 1; i < PART_LEN; i++) {
ef[0][i] = fft[2 * i];
ef[1][i] = fft[2 * i + 1];
}
if (aec->metricsMode == 1) {
// Note that the first PART_LEN samples in fft (before transformation) are
// zero. Hence, the scaling by two in UpdateLevel() should not be
// performed. That scaling is taken care of in UpdateMetrics() instead.
UpdateLevel(&aec->linoutlevel, ef);
}
// Scale error signal inversely with far power.
WebRtcAec_ScaleErrorSignal(aec, ef);
WebRtcAec_FilterAdaptation(aec, fft, ef);
NonLinearProcessing(aec, output, outputH);
if (aec->metricsMode == 1) {
// Update power levels and echo metrics
UpdateLevel(&aec->farlevel, (float (*)[PART_LEN1]) xf_ptr);
UpdateLevel(&aec->nearlevel, df);
UpdateMetrics(aec);
}
// Store the output block.
WebRtc_WriteBuffer(aec->outFrBuf, output, PART_LEN);
// For H band
if (aec->sampFreq == 32000) {
WebRtc_WriteBuffer(aec->outFrBufH, outputH, PART_LEN);
}
#ifdef WEBRTC_AEC_DEBUG_DUMP
{
int16_t eInt16[PART_LEN];
for (i = 0; i < PART_LEN; i++) {
eInt16[i] = (int16_t)WEBRTC_SPL_SAT(WEBRTC_SPL_WORD16_MAX, e[i],
WEBRTC_SPL_WORD16_MIN);
}
(void)fwrite(eInt16, sizeof(int16_t), PART_LEN, aec->outLinearFile);
(void)fwrite(output, sizeof(int16_t), PART_LEN, aec->outFile);
}
#endif
}