/// <summary>
/// Callback handler for intensity stereo info
/// </summary>
/// <param name="bits">
/// Speex bits buffer.
/// </param>
public void init(Bits bits)
{
float sign = 1;
int tmp;
if (bits.UnPack(1) != 0)
{
sign = -1;
}
tmp = bits.UnPack(5);
balance = (float)Math.Exp(sign * .25f * tmp);
tmp = bits.UnPack(2);
e_ratio = e_ratio_quant[tmp];
}
/// <summary>
/// User in-band request (submode=13).
/// </summary>
/// <param name="bits"></param>
public void userInbandRequest(Bits bits)
{
try
{
int req_size = bits.UnPack(4);
bits.Advance(5 + 8 * req_size);
}
catch (Exception e)
{
throw new SpeexException(e.Message);
}
}
/// <summary>
/// Decode the given input bits.
/// </summary>
/// <param name="bits">Speex bits buffer.</param>
/// <param name="vout">the decoded mono audio frame.</param>
/// <returns>
/// 1 if a terminator was found, 0 if not.
/// </returns>
public int Decode(Bits bits, float[] vout)
{
int i, sub, wideband, ret;
float[] low_pi_gain, low_exc, low_innov;
/* Decode the low-band */
ret = lowdec.Decode(bits, x0d);
if (ret != 0)
{
return(ret);
}
bool dtx = lowdec.Dtx;
if (bits == null)
{
decodeLost(vout, dtx);
return(0);
}
/* Check "wideband bit" */
wideband = bits.Peek();
if (wideband != 0)
{
/*Regular wideband frame, read the submode*/
wideband = bits.UnPack(1);
submodeID = bits.UnPack(3);
}
else
{
/* was a narrowband frame, set "null submode"*/
submodeID = 0;
}
for (i = 0; i < frameSize; i++)
{
excBuf[i] = 0;
}
if (submodes[submodeID] == null)
{
if (dtx)
{
decodeLost(vout, true);
return(0);
}
for (i = 0; i < frameSize; i++)
{
excBuf[i] = VERY_SMALL;
}
first = 1;
Filters.iir_mem2(excBuf, excIdx, interp_qlpc, high, 0, frameSize,
lpcSize, mem_sp);
filters.fir_mem_up(x0d, Codebook.h0, y0, fullFrameSize, QMF_ORDER, g0_mem);
filters.fir_mem_up(high, Codebook.h1, y1, fullFrameSize, QMF_ORDER, g1_mem);
for (i = 0; i < fullFrameSize; i++)
{
vout[i] = 2 * (y0[i] - y1[i]);
}
return(0);
}
low_pi_gain = lowdec.PitchGain;
low_exc = lowdec.Excitation;
low_innov = lowdec.Innovation;
submodes[submodeID].lsqQuant.unquant(qlsp, lpcSize, bits);
if (first != 0)
{
for (i = 0; i < lpcSize; i++)
{
old_qlsp[i] = qlsp[i];
}
}
for (sub = 0; sub < nbSubframes; sub++)
{
float tmp, filter_ratio, el = 0.0f, rl = 0.0f, rh = 0.0f;
int subIdx = subframeSize * sub;
/* LSP interpolation */
tmp = (1.0f + sub) / nbSubframes;
for (i = 0; i < lpcSize; i++)
{
interp_qlsp[i] = (1 - tmp) * old_qlsp[i] + tmp * qlsp[i];
}
Lsp.enforce_margin(interp_qlsp, lpcSize, .05f);
/* LSPs to x-domain */
for (i = 0; i < lpcSize; i++)
{
interp_qlsp[i] = (float)Math.Cos(interp_qlsp[i]);
}
/* LSP to LPC */
m_lsp.lsp2lpc(interp_qlsp, interp_qlpc, lpcSize);
if (enhanced)
{
float k1, k2, k3;
k1 = submodes[submodeID].lpc_enh_k1;
k2 = submodes[submodeID].lpc_enh_k2;
k3 = k1 - k2;
Filters.bw_lpc(k1, interp_qlpc, awk1, lpcSize);
Filters.bw_lpc(k2, interp_qlpc, awk2, lpcSize);
Filters.bw_lpc(k3, interp_qlpc, awk3, lpcSize);
}
/* Calculate reponse ratio between low & high filter in band middle (4000 Hz) */
tmp = 1;
pi_gain[sub] = 0;
for (i = 0; i <= lpcSize; i++)
{
rh += tmp * interp_qlpc[i];
tmp = -tmp;
pi_gain[sub] += interp_qlpc[i];
}
rl = low_pi_gain[sub];
rl = 1 / (Math.Abs(rl) + .01f);
rh = 1 / (Math.Abs(rh) + .01f);
filter_ratio = Math.Abs(.01f + rh) / (.01f + Math.Abs(rl));
/* reset excitation buffer */
for (i = subIdx; i < subIdx + subframeSize; i++)
{
excBuf[i] = 0;
}
if (submodes[submodeID].innovation == null)
{
float g;
int quant;
quant = bits.UnPack(5);
g = (float)Math.Exp(((double)quant - 10) / 8.0);
g /= filter_ratio;
/* High-band excitation using the low-band excitation and a gain */
for (i = subIdx; i < subIdx + subframeSize; i++)
{
excBuf[i] = foldingGain * g * low_innov[i];
}
}
else
{
float gc, scale;
int qgc = bits.UnPack(4);
for (i = subIdx; i < subIdx + subframeSize; i++)
{
el += low_exc[i] * low_exc[i];
}
gc = (float)Math.Exp((1 / 3.7f) * qgc - 2);
scale = gc * (float)Math.Sqrt(1 + el) / filter_ratio;
submodes[submodeID].innovation.UnQuant(excBuf, subIdx, subframeSize, bits);
for (i = subIdx; i < subIdx + subframeSize; i++)
{
excBuf[i] *= scale;
}
if (submodes[submodeID].double_codebook != 0)
{
for (i = 0; i < subframeSize; i++)
{
innov2[i] = 0;
}
submodes[submodeID].innovation.UnQuant(innov2, 0, subframeSize, bits);
for (i = 0; i < subframeSize; i++)
{
innov2[i] *= scale * (1 / 2.5f);
}
for (i = 0; i < subframeSize; i++)
{
excBuf[subIdx + i] += innov2[i];
}
}
}
for (i = subIdx; i < subIdx + subframeSize; i++)
{
high[i] = excBuf[i];
}
if (enhanced)
{
/* Use enhanced LPC filter */
Filters.filter_mem2(high, subIdx, awk2, awk1, subframeSize,
lpcSize, mem_sp, lpcSize);
Filters.filter_mem2(high, subIdx, awk3, interp_qlpc, subframeSize,
lpcSize, mem_sp, 0);
}
else
{
/* Use regular filter */
for (i = 0; i < lpcSize; i++)
{
mem_sp[lpcSize + i] = 0;
}
Filters.iir_mem2(high, subIdx, interp_qlpc, high, subIdx,
subframeSize, lpcSize, mem_sp);
}
}
filters.fir_mem_up(x0d, Codebook.h0, y0, fullFrameSize, QMF_ORDER, g0_mem);
filters.fir_mem_up(high, Codebook.h1, y1, fullFrameSize, QMF_ORDER, g1_mem);
for (i = 0; i < fullFrameSize; i++)
{
vout[i] = 2 * (y0[i] - y1[i]);
}
for (i = 0; i < lpcSize; i++)
{
old_qlsp[i] = qlsp[i];
}
first = 0;
return(0);
}
/// <summary>
/// Speex in-band request (submode=14).
/// </summary>
/// <param name="bits"></param>
public void speexInbandRequest(Bits bits)
{
int code = bits.UnPack(4);
switch (code)
{
case 0: // asks the decoder to set perceptual enhancment off (0) or on (1)
bits.Advance(1);
break;
case 1: // asks (if 1) the encoder to be less "aggressive" due to high packet loss
bits.Advance(1);
break;
case 2: // asks the encoder to switch to mode N
bits.Advance(4);
break;
case 3: // asks the encoder to switch to mode N for low-band
bits.Advance(4);
break;
case 4: // asks the encoder to switch to mode N for high-band
bits.Advance(4);
break;
case 5: // asks the encoder to switch to quality N for VBR
bits.Advance(4);
break;
case 6: // request acknowledgement (0=no, 1=all, 2=only for inband data)
bits.Advance(4);
break;
case 7: // asks the encoder to set CBR(0), VAD(1), DTX(3), VBR(5), VBR+DTX(7)
bits.Advance(4);
break;
case 8: // transmit (8-bit) character to the other end
bits.Advance(8);
break;
case 9: // intensity stereo information
// setup the stereo decoder; to skip: tmp = bits.unpack(8); break;
_stereo.init(bits); // read 8 bits
break;
case 10: // announce maximum bit-rate acceptable (N in byets/second)
bits.Advance(16);
break;
case 11: // reserved
bits.Advance(16);
break;
case 12: // Acknowledge receiving packet N
bits.Advance(32);
break;
case 13: // reserved
bits.Advance(32);
break;
case 14: // reserved
bits.Advance(64);
break;
case 15: // reserved
bits.Advance(64);
break;
default:
break;
}
}
/// <summary>
/// Decode the given input bits.
/// </summary>
/// <param name="bits">Speex bits buffer.</param>
/// <param name="vout">the decoded mono audio frame.</param>
/// <returns>1 if a terminator was found, 0 if not.</returns>
public int Decode(Bits bits, float[] vout)
{
int i, sub, pitch, ol_pitch = 0, m;
float[] pitch_gain = new float[3];
float ol_gain = 0.0f, ol_pitch_coef = 0.0f;
int best_pitch = 40;
float best_pitch_gain = 0;
float pitch_average = 0;
if (bits == null && dtx_enabled != 0)
{
submodeID = 0;
}
else
{
if (bits == null)
{
decodeLost(vout);
return(0);
}
do
{
if (bits.UnPack(1) != 0)
{
m = bits.UnPack(SbCodec.SB_SUBMODE_BITS);
int advance = SbCodec.SB_FRAME_SIZE[m];
if (advance < 0)
{
throw new Exception("Invalid sideband mode encountered (1st sideband): " + m);
}
advance -= (SbCodec.SB_SUBMODE_BITS + 1);
bits.Advance(advance);
if (bits.UnPack(1) != 0)
{ /* Skip ultra-wideband block (for compatibility) */
/* Get the sub-mode that was used */
m = bits.UnPack(SbCodec.SB_SUBMODE_BITS);
advance = SbCodec.SB_FRAME_SIZE[m];
if (advance < 0)
{
throw new Exception("Invalid sideband mode encountered. (2nd sideband): " + m);
//return -2;
}
advance -= (SbCodec.SB_SUBMODE_BITS + 1);
bits.Advance(advance);
if (bits.UnPack(1) != 0)
{ /* Sanity check */
throw new Exception("More than two sideband layers found");
//return -2;
}
}
}
m = bits.UnPack(NB_SUBMODE_BITS);
if (m == 15)
{ /* We found a terminator */
return(1);
}
else if (m == 14)
{ /* Speex in-band request */
inband.speexInbandRequest(bits);
}
else if (m == 13)
{ /* User in-band request */
inband.userInbandRequest(bits);
}
else if (m > 8)
{ /* Invalid mode */
throw new Exception("Invalid mode encountered: " + m);
//return -2;
}
}while (m > 8);
submodeID = m;
}
/* Shift all buffers by one frame */
System.Array.Copy(frmBuf, frameSize, frmBuf, 0, bufSize - frameSize);
System.Array.Copy(excBuf, frameSize, excBuf, 0, bufSize - frameSize);
if (submodes[submodeID] == null)
{
Filters.bw_lpc(.93f, interp_qlpc, lpc, 10);
float innov_gain = 0;
for (i = 0; i < frameSize; i++)
{
innov_gain += innov[i] * innov[i];
}
innov_gain = (float)Math.Sqrt(innov_gain / frameSize);
for (i = excIdx; i < excIdx + frameSize; i++)
{
excBuf[i] = 3 * innov_gain * ((float)random.NextDouble() - .5f);
}
first = 1;
/* Final signal synthesis from excitation */
Filters.iir_mem2(excBuf, excIdx, lpc, frmBuf, frmIdx, frameSize, lpcSize, mem_sp);
vout[0] = frmBuf[frmIdx] + preemph * pre_mem;
for (i = 1; i < frameSize; i++)
{
vout[i] = frmBuf[frmIdx + i] + preemph * vout[i - 1];
}
pre_mem = vout[frameSize - 1];
count_lost = 0;
return(0);
}
/* Unquantize LSPs */
submodes[submodeID].lsqQuant.unquant(qlsp, lpcSize, bits);
if (count_lost != 0)
{
float lsp_dist = 0, fact;
for (i = 0; i < lpcSize; i++)
{
lsp_dist += Math.Abs(old_qlsp[i] - qlsp[i]);
}
fact = (float)(.6 * Math.Exp(-.2 * lsp_dist));
for (i = 0; i < 2 * lpcSize; i++)
{
mem_sp[i] *= fact;
}
}
/* Handle first frame and lost-packet case */
if (first != 0 || count_lost != 0)
{
for (i = 0; i < lpcSize; i++)
{
old_qlsp[i] = qlsp[i];
}
}
/* Get open-loop pitch estimation for low bit-rate pitch coding */
if (submodes[submodeID].lbr_pitch != -1)
{
ol_pitch = min_pitch + bits.UnPack(7);
}
if (submodes[submodeID].forced_pitch_gain != 0)
{
int quant = bits.UnPack(4);
ol_pitch_coef = 0.066667f * quant;
}
/* Get global excitation gain */
int qe = bits.UnPack(5);
ol_gain = (float)Math.Exp(qe / 3.5);
/* unpacks unused dtx bits */
if (submodeID == 1)
{
int extra = bits.UnPack(4);
if (extra == 15)
{
dtx_enabled = 1;
}
else
{
dtx_enabled = 0;
}
}
if (submodeID > 1)
{
dtx_enabled = 0;
}
/*Loop on subframes */
for (sub = 0; sub < nbSubframes; sub++)
{
int offset, spIdx, extIdx;
float tmp;
/* Offset relative to start of frame */
offset = subframeSize * sub;
/* Original signal */
spIdx = frmIdx + offset;
/* Excitation */
extIdx = excIdx + offset;
/* LSP interpolation (quantized and unquantized) */
tmp = (1.0f + sub) / nbSubframes;
for (i = 0; i < lpcSize; i++)
{
interp_qlsp[i] = (1 - tmp) * old_qlsp[i] + tmp * qlsp[i];
}
/* Make sure the LSP's are stable */
Lsp.enforce_margin(interp_qlsp, lpcSize, .002f);
/* Compute interpolated LPCs (unquantized) */
for (i = 0; i < lpcSize; i++)
{
interp_qlsp[i] = (float)Math.Cos(interp_qlsp[i]);
}
m_lsp.lsp2lpc(interp_qlsp, interp_qlpc, lpcSize);
/* Compute enhanced synthesis filter */
if (enhanced)
{
float r = .9f;
float k1, k2, k3;
k1 = submodes[submodeID].lpc_enh_k1;
k2 = submodes[submodeID].lpc_enh_k2;
k3 = (1 - (1 - r * k1) / (1 - r * k2)) / r;
Filters.bw_lpc(k1, interp_qlpc, awk1, lpcSize);
Filters.bw_lpc(k2, interp_qlpc, awk2, lpcSize);
Filters.bw_lpc(k3, interp_qlpc, awk3, lpcSize);
}
/* Compute analysis filter at w=pi */
tmp = 1;
pi_gain[sub] = 0;
for (i = 0; i <= lpcSize; i++)
{
pi_gain[sub] += tmp * interp_qlpc[i];
tmp = -tmp;
}
/* Reset excitation */
for (i = 0; i < subframeSize; i++)
{
excBuf[extIdx + i] = 0;
}
/*Adaptive codebook contribution*/
int pit_min, pit_max;
/* Handle pitch constraints if any */
if (submodes[submodeID].lbr_pitch != -1)
{
int margin = submodes[submodeID].lbr_pitch;
if (margin != 0)
{
pit_min = ol_pitch - margin + 1;
if (pit_min < min_pitch)
{
pit_min = min_pitch;
}
pit_max = ol_pitch + margin;
if (pit_max > max_pitch)
{
pit_max = max_pitch;
}
}
else
{
pit_min = pit_max = ol_pitch;
}
}
else
{
pit_min = min_pitch;
pit_max = max_pitch;
}
/* Pitch synthesis */
pitch = submodes[submodeID].ltp.UnQuant(excBuf, extIdx, pit_min, ol_pitch_coef,
subframeSize, pitch_gain, bits,
count_lost, offset, last_pitch_gain);
/* If we had lost frames, check energy of last received frame */
if (count_lost != 0 && ol_gain < last_ol_gain)
{
float fact = ol_gain / (last_ol_gain + 1);
for (i = 0; i < subframeSize; i++)
{
excBuf[excIdx + i] *= fact;
}
}
tmp = Math.Abs(pitch_gain[0] + pitch_gain[1] + pitch_gain[2]);
tmp = Math.Abs(pitch_gain[1]);
if (pitch_gain[0] > 0)
{
tmp += pitch_gain[0];
}
else
{
tmp -= .5f * pitch_gain[0];
}
if (pitch_gain[2] > 0)
{
tmp += pitch_gain[2];
}
else
{
tmp -= .5f * pitch_gain[0];
}
pitch_average += tmp;
if (tmp > best_pitch_gain)
{
best_pitch = pitch;
best_pitch_gain = tmp;
}
/* Unquantize the innovation */
int q_energy, ivi = sub * subframeSize;
float ener;
for (i = ivi; i < ivi + subframeSize; i++)
{
innov[i] = 0.0f;
}
/* Decode sub-frame gain correction */
if (submodes[submodeID].have_subframe_gain == 3)
{
q_energy = bits.UnPack(3);
ener = (float)(ol_gain * Math.Exp(exc_gain_quant_scal3[q_energy]));
}
else if (submodes[submodeID].have_subframe_gain == 1)
{
q_energy = bits.UnPack(1);
ener = (float)(ol_gain * Math.Exp(exc_gain_quant_scal1[q_energy]));
}
else
{
ener = ol_gain;
}
if (submodes[submodeID].innovation != null)
{
/* Fixed codebook contribution */
submodes[submodeID].innovation.UnQuant(innov, ivi, subframeSize, bits);
}
/* De-normalize innovation and update excitation */
for (i = ivi; i < ivi + subframeSize; i++)
{
innov[i] *= ener;
}
/* Vocoder mode */
if (submodeID == 1)
{
float g = ol_pitch_coef;
for (i = 0; i < subframeSize; i++)
{
excBuf[extIdx + i] = 0;
}
while (voc_offset < subframeSize)
{
if (voc_offset >= 0)
{
excBuf[extIdx + voc_offset] = (float)Math.Sqrt(1.0f * ol_pitch);
}
voc_offset += ol_pitch;
}
voc_offset -= subframeSize;
g = .5f + 2 * (g - .6f);
if (g < 0)
{
g = 0;
}
if (g > 1)
{
g = 1;
}
for (i = 0; i < subframeSize; i++)
{
float itmp = excBuf[extIdx + i];
excBuf[extIdx + i] = .8f * g * excBuf[extIdx + i] * ol_gain + .6f * g * voc_m1 * ol_gain + .5f * g * innov[ivi + i] - .5f * g * voc_m2 + (1 - g) * innov[ivi + i];
voc_m1 = itmp;
voc_m2 = innov[ivi + i];
voc_mean = .95f * voc_mean + .05f * excBuf[extIdx + i];
excBuf[extIdx + i] -= voc_mean;
}
}
else
{
for (i = 0; i < subframeSize; i++)
{
excBuf[extIdx + i] += innov[ivi + i];
}
}
/* Decode second codebook (only for some modes) */
if (submodes[submodeID].double_codebook != 0)
{
for (i = 0; i < subframeSize; i++)
{
innov2[i] = 0;
}
submodes[submodeID].innovation.UnQuant(innov2, 0, subframeSize, bits);
for (i = 0; i < subframeSize; i++)
{
innov2[i] *= ener * (1f / 2.2f);
}
for (i = 0; i < subframeSize; i++)
{
excBuf[extIdx + i] += innov2[i];
}
}
for (i = 0; i < subframeSize; i++)
{
frmBuf[spIdx + i] = excBuf[extIdx + i];
}
/* Signal synthesis */
if (enhanced && submodes[submodeID].comb_gain > 0)
{
filters.comb_filter(excBuf, extIdx, frmBuf, spIdx, subframeSize,
pitch, pitch_gain, submodes[submodeID].comb_gain);
}
if (enhanced)
{
/* Use enhanced LPC filter */
Filters.filter_mem2(frmBuf, spIdx, awk2, awk1, subframeSize, lpcSize, mem_sp, lpcSize);
Filters.filter_mem2(frmBuf, spIdx, awk3, interp_qlpc, subframeSize, lpcSize, mem_sp, 0);
}
else
{
/* Use regular filter */
for (i = 0; i < lpcSize; i++)
{
mem_sp[lpcSize + i] = 0;
}
Filters.iir_mem2(frmBuf, spIdx, interp_qlpc, frmBuf, spIdx, subframeSize, lpcSize, mem_sp);
}
}
/*Copy output signal*/
vout[0] = frmBuf[frmIdx] + preemph * pre_mem;
for (i = 1; i < frameSize; i++)
{
vout[i] = frmBuf[frmIdx + i] + preemph * vout[i - 1];
}
pre_mem = vout[frameSize - 1];
/* Store the LSPs for interpolation in the next frame */
for (i = 0; i < lpcSize; i++)
{
old_qlsp[i] = qlsp[i];
}
/* The next frame will not be the first (Duh!) */
first = 0;
count_lost = 0;
last_pitch = best_pitch;
last_pitch_gain = .25f * pitch_average;
pitch_gain_buf[pitch_gain_buf_idx++] = last_pitch_gain;
if (pitch_gain_buf_idx > 2) /* rollover */
{
pitch_gain_buf_idx = 0;
}
last_ol_gain = ol_gain;
return(0);
}
/// <summary>
/// Long Term Prediction Unquantification (3Tap).
/// </summary>
/// <param name="exc"> Excitation</param>
/// <param name="es">Excitation offset</param>
/// <param name="start">Smallest pitch value allowed</param>
/// <param name="pitch_coef"> Voicing (pitch) coefficient</param>
/// <param name="nsf">Number of samples in subframe</param>
/// <param name="gain_val"></param>
/// <param name="bits">Speex bits buffer.</param>
/// <param name="count_lost"></param>
/// <param name="subframe_offset"></param>
/// <param name="last_pitch_gain"></param>
/// <returns>pitch</returns>
public override sealed int UnQuant(float[] exc, int es, int start, float pitch_coef, int nsf, float[] gain_val, Bits bits, int count_lost, int subframe_offset, float last_pitch_gain)
{
int i, pitch, gain_index;
pitch = bits.UnPack(pitch_bits);
pitch += start;
gain_index = bits.UnPack(gain_bits);
gain[0] = 0.015625f * (float)gain_cdbk[gain_index * 3] + .5f;
gain[1] = 0.015625f * (float)gain_cdbk[gain_index * 3 + 1] + .5f;
gain[2] = 0.015625f * (float)gain_cdbk[gain_index * 3 + 2] + .5f;
if (count_lost != 0 && pitch > subframe_offset)
{
float gain_sum = Math.Abs(gain[1]);
float tmp = count_lost < 4 ? last_pitch_gain : 0.4f * last_pitch_gain;
if (tmp > .95f)
{
tmp = .95f;
}
if (gain[0] > 0)
{
gain_sum += gain[0];
}
else
{
gain_sum -= .5f * gain[0];
}
if (gain[2] > 0)
{
gain_sum += gain[2];
}
else
{
gain_sum -= .5f * gain[0];
}
if (gain_sum > tmp)
{
float fact = tmp / gain_sum;
for (i = 0; i < 3; i++)
{
gain[i] *= fact;
}
}
}
gain_val[0] = gain[0];
gain_val[1] = gain[1];
gain_val[2] = gain[2];
for (i = 0; i < 3; i++)
{
int j, tmp1, tmp2, pp = pitch + 1 - i;
tmp1 = nsf;
if (tmp1 > pp)
{
tmp1 = pp;
}
tmp2 = nsf;
if (tmp2 > pp + pitch)
{
tmp2 = pp + pitch;
}
for (j = 0; j < tmp1; j++)
{
e[i][j] = exc[es + j - pp];
}
for (j = tmp1; j < tmp2; j++)
{
e[i][j] = exc[es + j - pp - pitch];
}
for (j = tmp2; j < nsf; j++)
{
e[i][j] = 0;
}
}
for (i = 0; i < nsf; i++)
{
exc[es + i] = gain[0] * e[2][i] + gain[1] * e[1][i] + gain[2] * e[0][i];
}
return(pitch);
}
