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;
}
void WebRtcAec_BufferFarendPartition(aec_t *aec, const float* farend) {
float fft[PART_LEN2];
float xf[2][PART_LEN1];
// Check if the buffer is full, and in that case flush the oldest data.
if (WebRtc_available_write(aec->far_buf) < 1) {
WebRtcAec_MoveFarReadPtr(aec, 1);
}
// Convert far-end partition to the frequency domain without windowing.
memcpy(fft, farend, sizeof(float) * PART_LEN2);
TimeToFrequency(fft, xf, 0);
WebRtc_WriteBuffer(aec->far_buf, &xf[0][0], 1);
// Convert far-end partition to the frequency domain with windowing.
memcpy(fft, farend, sizeof(float) * PART_LEN2);
TimeToFrequency(fft, xf, 1);
WebRtc_WriteBuffer(aec->far_buf_windowed, &xf[0][0], 1);
}
// 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;
}
// 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 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
}