/// <summary>
/// Encode the given input signal.
/// </summary>
/// <param name="bits">Speex bits buffer.</param>
/// <param name="vin">the raw mono audio frame to encode</param>
/// <returns>1 if successful.</returns>
public int Encode(Bits bits, float[] vin)
{
int i;
float[] mem, innov, syn_resp;
float[] low_pi_gain, low_exc, low_innov;
int dtx;
/* Compute the two sub-bands by filtering with h0 and h1*/
Filters.qmf_decomp(vin, h0, x0d, x1d, fullFrameSize, QMF_ORDER, h0_mem);
/* Encode the narrowband part*/
lowenc.Encode(bits, x0d);
/* High-band buffering / sync with low band */
for (i = 0; i < windowSize - frameSize; i++)
{
high[i] = high[frameSize + i];
}
for (i = 0; i < frameSize; i++)
{
high[windowSize - frameSize + i] = x1d[i];
}
Array.Copy(excBuf, frameSize, excBuf, 0, bufSize - frameSize);
low_pi_gain = lowenc.PitchGain;
low_exc = lowenc.Excitation;
low_innov = lowenc.Innovation;
int low_mode = lowenc.Mode;
if (low_mode == 0)
{
dtx = 1;
}
else
{
dtx = 0;
}
/* Start encoding the high-band */
for (i = 0; i < windowSize; i++)
{
buf[i] = high[i] * window[i];
}
/* Compute auto-correlation */
Lpc.autocorr(buf, autocorr, lpcSize + 1, windowSize);
autocorr[0] += 1; /* prevents NANs */
autocorr[0] *= lpc_floor; /* Noise floor in auto-correlation domain */
/* Lag windowing: equivalent to filtering in the power-spectrum domain */
for (i = 0; i < lpcSize + 1; i++)
{
autocorr[i] *= lagWindow[i];
}
/* Levinson-Durbin */
Lpc.wld(lpc, autocorr, rc, lpcSize); // tmperr
Array.Copy(lpc, 0, lpc, 1, lpcSize);
lpc[0] = 1;
/* LPC to LSPs (x-domain) transform */
int roots = Lsp.lpc2lsp(lpc, lpcSize, lsp, 15, 0.2f);
if (roots != lpcSize)
{
roots = Lsp.lpc2lsp(lpc, lpcSize, lsp, 11, 0.02f);
if (roots != lpcSize)
{
/*If we can't find all LSP's, do some damage control and use a flat filter*/
for (i = 0; i < lpcSize; i++)
{
lsp[i] = (float)Math.Cos(Math.PI * ((float)(i + 1)) / (lpcSize + 1));
}
}
}
/* x-domain to angle domain*/
for (i = 0; i < lpcSize; i++)
{
lsp[i] = (float)Math.Acos(lsp[i]);
}
float lsp_dist = 0;
for (i = 0; i < lpcSize; i++)
{
lsp_dist += (old_lsp[i] - lsp[i]) * (old_lsp[i] - lsp[i]);
}
/*VBR stuff*/
if ((vbr_enabled != 0 || vad_enabled != 0) && dtx == 0)
{
float e_low = 0, e_high = 0;
float ratio;
if (abr_enabled != 0)
{
float qual_change = 0;
if (abr_drift2 * abr_drift > 0)
{
/* Only adapt if long-term and short-term drift are the same sign */
qual_change = -.00001f * abr_drift / (1 + abr_count);
if (qual_change > .1f)
{
qual_change = .1f;
}
if (qual_change < -.1f)
{
qual_change = -.1f;
}
}
vbr_quality += qual_change;
if (vbr_quality > 10)
{
vbr_quality = 10;
}
if (vbr_quality < 0)
{
vbr_quality = 0;
}
}
for (i = 0; i < frameSize; i++)
{
e_low += x0d[i] * x0d[i];
e_high += high[i] * high[i];
}
ratio = (float)Math.Log((1 + e_high) / (1 + e_low));
relative_quality = lowenc.RelativeQuality;
if (ratio < -4)
{
ratio = -4;
}
if (ratio > 2)
{
ratio = 2;
}
/*if (ratio>-2)*/
if (vbr_enabled != 0)
{
int modeid;
modeid = nb_modes - 1;
relative_quality += 1.0f * (ratio + 2f);
if (relative_quality < -1)
{
relative_quality = -1;
}
while (modeid != 0)
{
int v1;
float thresh;
v1 = (int)Math.Floor(vbr_quality);
if (v1 == 10)
{
thresh = NSpeex.Vbr.hb_thresh[modeid][v1];
}
else
{
thresh = (vbr_quality - v1) * NSpeex.Vbr.hb_thresh[modeid][v1 + 1] + (1 + v1 - vbr_quality) * NSpeex.Vbr.hb_thresh[modeid][v1];
}
if (relative_quality >= thresh)
{
break;
}
modeid--;
}
Mode = modeid;
if (abr_enabled != 0)
{
int bitrate = BitRate;
abr_drift += (bitrate - abr_enabled);
abr_drift2 = .95f * abr_drift2 + .05f * (bitrate - abr_enabled);
abr_count += 1.0f;
}
}
else
{
/* VAD only */
int modeid;
if (relative_quality < 2.0)
{
modeid = 1;
}
else
{
modeid = submodeSelect;
}
/*speex_encoder_ctl(state, SPEEX_SET_MODE, &mode);*/
submodeID = modeid;
}
}
bits.Pack(1, 1);
if (dtx != 0)
{
bits.Pack(0, SB_SUBMODE_BITS);
}
else
{
bits.Pack(submodeID, SB_SUBMODE_BITS);
}
/* If null mode (no transmission), just set a couple things to zero*/
if (dtx != 0 || submodes[submodeID] == null)
{
for (i = 0; i < frameSize; i++)
{
excBuf[excIdx + i] = swBuf[i] = VERY_SMALL;
}
for (i = 0; i < lpcSize; i++)
{
mem_sw[i] = 0;
}
first = 1;
/* Final signal synthesis from excitation */
Filters.iir_mem2(excBuf, excIdx, interp_qlpc, high, 0, subframeSize, lpcSize, mem_sp);
/* Reconstruct the original */
filters.fir_mem_up(x0d, h0, y0, fullFrameSize, QMF_ORDER, g0_mem);
filters.fir_mem_up(high, h1, y1, fullFrameSize, QMF_ORDER, g1_mem);
for (i = 0; i < fullFrameSize; i++)
{
vin[i] = 2 * (y0[i] - y1[i]);
}
if (dtx != 0)
{
return(0);
}
else
{
return(1);
}
}
/* LSP quantization */
submodes[submodeID].lsqQuant.quant(lsp, qlsp, lpcSize, bits);
if (first != 0)
{
for (i = 0; i < lpcSize; i++)
{
old_lsp[i] = lsp[i];
}
for (i = 0; i < lpcSize; i++)
{
old_qlsp[i] = qlsp[i];
}
}
mem = new float[lpcSize];
syn_resp = new float[subframeSize];
innov = new float[subframeSize];
for (int sub = 0; sub < nbSubframes; sub++)
{
float tmp, filter_ratio;
int exc, sp, sw, resp;
int offset;
float rl, rh, eh = 0, el = 0;
int fold;
offset = subframeSize * sub;
sp = offset;
exc = excIdx + offset;
resp = offset;
sw = offset;
/* LSP interpolation (quantized and unquantized) */
tmp = (1.0f + sub) / nbSubframes;
for (i = 0; i < lpcSize; i++)
{
interp_lsp[i] = (1 - tmp) * old_lsp[i] + tmp * lsp[i];
}
for (i = 0; i < lpcSize; i++)
{
interp_qlsp[i] = (1 - tmp) * old_qlsp[i] + tmp * qlsp[i];
}
Lsp.enforce_margin(interp_lsp, lpcSize, .05f);
Lsp.enforce_margin(interp_qlsp, lpcSize, .05f);
/* Compute interpolated LPCs (quantized and unquantized) */
for (i = 0; i < lpcSize; i++)
{
interp_lsp[i] = (float)Math.Cos(interp_lsp[i]);
}
for (i = 0; i < lpcSize; i++)
{
interp_qlsp[i] = (float)Math.Cos(interp_qlsp[i]);
}
m_lsp.lsp2lpc(interp_lsp, interp_lpc, lpcSize);
m_lsp.lsp2lpc(interp_qlsp, interp_qlpc, lpcSize);
Filters.bw_lpc(gamma1, interp_lpc, bw_lpc1, lpcSize);
Filters.bw_lpc(gamma2, interp_lpc, bw_lpc2, lpcSize);
/* Compute mid-band (4000 Hz for wideband) response of low-band and high-band
* filters */
rl = rh = 0;
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);
/* Compute ratio, will help predict the gain */
filter_ratio = Math.Abs(.01f + rh) / (.01f + Math.Abs(rl));
fold = filter_ratio < 5 ? 1 : 0;
/*printf ("filter_ratio %f\n", filter_ratio);*/
fold = 0;
/* Compute "real excitation" */
Filters.fir_mem2(high, sp, interp_qlpc, excBuf, exc, subframeSize, lpcSize, mem_sp2);
/* Compute energy of low-band and high-band excitation */
for (i = 0; i < subframeSize; i++)
{
eh += excBuf[exc + i] * excBuf[exc + i];
}
if (submodes[submodeID].innovation == null)
{
float g;
/*speex_bits_pack(bits, 1, 1);*/
for (i = 0; i < subframeSize; i++)
{
el += low_innov[offset + i] * low_innov[offset + i];
}
/* Gain to use if we want to use the low-band excitation for high-band */
g = eh / (.01f + el);
g = (float)Math.Sqrt(g);
g *= filter_ratio;
int quant = (int)Math.Floor(.5 + 10 + 8.0 * Math.Log((g + .0001)));
/*speex_warning_int("tata", quant);*/
if (quant < 0)
{
quant = 0;
}
if (quant > 31)
{
quant = 31;
}
bits.Pack(quant, 5);
g = (float)(.1 * Math.Exp(quant / 9.4));
g /= filter_ratio;
}
else
{
float gc, scale, scale_1;
for (i = 0; i < subframeSize; i++)
{
el += low_exc[offset + i] * low_exc[offset + i];
}
gc = (float)(Math.Sqrt(1 + eh) * filter_ratio / Math.Sqrt((1 + el) * subframeSize));
{
int qgc = (int)Math.Floor(.5 + 3.7 * (Math.Log(gc) + 2));
if (qgc < 0)
{
qgc = 0;
}
if (qgc > 15)
{
qgc = 15;
}
bits.Pack(qgc, 4);
gc = (float)Math.Exp((1 / 3.7) * qgc - 2);
}
scale = gc * (float)Math.Sqrt(1 + el) / filter_ratio;
scale_1 = 1 / scale;
for (i = 0; i < subframeSize; i++)
{
excBuf[exc + i] = 0;
}
excBuf[exc] = 1;
Filters.syn_percep_zero(excBuf, exc, interp_qlpc, bw_lpc1, bw_lpc2, syn_resp, subframeSize, lpcSize);
/* Reset excitation */
for (i = 0; i < subframeSize; i++)
{
excBuf[exc + i] = 0;
}
/* Compute zero response (ringing) of A(z/g1) / ( A(z/g2) * Aq(z) ) */
for (i = 0; i < lpcSize; i++)
{
mem[i] = mem_sp[i];
}
Filters.iir_mem2(excBuf, exc, interp_qlpc, excBuf, exc, subframeSize, lpcSize, mem);
for (i = 0; i < lpcSize; i++)
{
mem[i] = mem_sw[i];
}
Filters.filter_mem2(excBuf, exc, bw_lpc1, bw_lpc2, res, resp, subframeSize, lpcSize, mem, 0);
/* Compute weighted signal */
for (i = 0; i < lpcSize; i++)
{
mem[i] = mem_sw[i];
}
Filters.filter_mem2(high, sp, bw_lpc1, bw_lpc2, swBuf, sw, subframeSize, lpcSize, mem, 0);
/* Compute target signal */
for (i = 0; i < subframeSize; i++)
{
target[i] = swBuf[sw + i] - res[resp + i];
}
for (i = 0; i < subframeSize; i++)
{
excBuf[exc + i] = 0;
}
for (i = 0; i < subframeSize; i++)
{
target[i] *= scale_1;
}
/* Reset excitation */
for (i = 0; i < subframeSize; i++)
{
innov[i] = 0;
}
/*print_vec(target, st->subframeSize, "\ntarget");*/
submodes[submodeID].innovation.Quant(target, interp_qlpc, bw_lpc1, bw_lpc2,
lpcSize, subframeSize, innov, 0, syn_resp,
bits, (complexity + 1) >> 1);
/*print_vec(target, st->subframeSize, "after");*/
for (i = 0; i < subframeSize; i++)
{
excBuf[exc + i] += innov[i] * scale;
}
if (submodes[submodeID].double_codebook != 0)
{
float[] innov2 = new float[subframeSize];
for (i = 0; i < subframeSize; i++)
{
innov2[i] = 0;
}
for (i = 0; i < subframeSize; i++)
{
target[i] *= 2.5f;
}
submodes[submodeID].innovation.Quant(target, interp_qlpc, bw_lpc1, bw_lpc2,
lpcSize, subframeSize, innov2, 0, syn_resp,
bits, (complexity + 1) >> 1);
for (i = 0; i < subframeSize; i++)
{
innov2[i] *= scale * (1f / 2.5f);
}
for (i = 0; i < subframeSize; i++)
{
excBuf[exc + i] += innov2[i];
}
}
}
/*Keep the previous memory*/
for (i = 0; i < lpcSize; i++)
{
mem[i] = mem_sp[i];
}
/* Final signal synthesis from excitation */
Filters.iir_mem2(excBuf, exc, interp_qlpc, high, sp, subframeSize, lpcSize, mem_sp);
/* Compute weighted signal again, from synthesized speech (not sure it's the right thing) */
Filters.filter_mem2(high, sp, bw_lpc1, bw_lpc2, swBuf, sw, subframeSize, lpcSize, mem_sw, 0);
}
//#ifndef RELEASE
/* Reconstruct the original */
filters.fir_mem_up(x0d, h0, y0, fullFrameSize, QMF_ORDER, g0_mem);
filters.fir_mem_up(high, h1, y1, fullFrameSize, QMF_ORDER, g1_mem);
for (i = 0; i < fullFrameSize; i++)
{
vin[i] = 2 * (y0[i] - y1[i]);
}
//#endif
for (i = 0; i < lpcSize; i++)
{
old_lsp[i] = lsp[i];
}
for (i = 0; i < lpcSize; i++)
{
old_qlsp[i] = qlsp[i];
}
first = 0;
return(1);
}
/// <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);
}
public int Encode(Bits bits, float[] vin)
{
int i;
float[] res, target, mem;
float[] syn_resp;
float[] orig;
/* Copy new data in input buffer */
System.Array.Copy(frmBuf, frameSize, frmBuf, 0, bufSize - frameSize);
frmBuf[bufSize - frameSize] = vin[0] - preemph * pre_mem;
for (i = 1; i < frameSize; i++)
{
frmBuf[bufSize - frameSize + i] = vin[i] - preemph * vin[i - 1];
}
pre_mem = vin[frameSize - 1];
/* Move signals 1 frame towards the past */
System.Array.Copy(exc2Buf, frameSize, exc2Buf, 0, bufSize - frameSize);
System.Array.Copy(excBuf, frameSize, excBuf, 0, bufSize - frameSize);
System.Array.Copy(swBuf, frameSize, swBuf, 0, bufSize - frameSize);
/* Window for analysis */
for (i = 0; i < windowSize; i++)
{
buf2[i] = frmBuf[i + frmIdx] * window[i];
}
/* Compute auto-correlation */
Lpc.autocorr(buf2, autocorr, lpcSize + 1, windowSize);
autocorr[0] += 10; /* prevents NANs */
autocorr[0] *= lpc_floor; /* Noise floor in auto-correlation domain */
/* Lag windowing: equivalent to filtering in the power-spectrum domain */
for (i = 0; i < lpcSize + 1; i++)
{
autocorr[i] *= lagWindow[i];
}
/* Levinson-Durbin */
Lpc.wld(lpc, autocorr, rc, lpcSize); // tmperr
System.Array.Copy(lpc, 0, lpc, 1, lpcSize);
lpc[0] = 1;
/* LPC to LSPs (x-domain) transform */
int roots = Lsp.lpc2lsp(lpc, lpcSize, lsp, 15, 0.2f);
/* Check if we found all the roots */
if (roots == lpcSize)
{
/* LSP x-domain to angle domain*/
for (i = 0; i < lpcSize; i++)
{
lsp[i] = (float)Math.Acos(lsp[i]);
}
}
else
{
/* Search again if we can afford it */
if (complexity > 1)
{
roots = Lsp.lpc2lsp(lpc, lpcSize, lsp, 11, 0.05f);
}
if (roots == lpcSize)
{
/* LSP x-domain to angle domain*/
for (i = 0; i < lpcSize; i++)
{
lsp[i] = (float)Math.Acos(lsp[i]);
}
}
else
{
/*If we can't find all LSP's, do some damage control and use previous filter*/
for (i = 0; i < lpcSize; i++)
{
lsp[i] = old_lsp[i];
}
}
}
float lsp_dist = 0;
for (i = 0; i < lpcSize; i++)
{
lsp_dist += (old_lsp[i] - lsp[i]) * (old_lsp[i] - lsp[i]);
}
/* Whole frame analysis (open-loop estimation of pitch and excitation gain) */
float ol_gain;
int ol_pitch;
float ol_pitch_coef;
{
if (first != 0)
{
for (i = 0; i < lpcSize; i++)
{
interp_lsp[i] = lsp[i];
}
}
else
{
for (i = 0; i < lpcSize; i++)
{
interp_lsp[i] = .375f * old_lsp[i] + .625f * lsp[i];
}
}
Lsp.enforce_margin(interp_lsp, lpcSize, .002f);
/* Compute interpolated LPCs (unquantized) for whole frame*/
for (i = 0; i < lpcSize; i++)
{
interp_lsp[i] = (float)Math.Cos(interp_lsp[i]);
}
m_lsp.lsp2lpc(interp_lsp, interp_lpc, lpcSize);
/*Open-loop pitch*/
if (submodes[submodeID] == null ||
vbr_enabled != 0 || vad_enabled != 0 ||
submodes[submodeID].forced_pitch_gain != 0 ||
submodes[submodeID].lbr_pitch != -1)
{
int[] nol_pitch = new int[6];
float[] nol_pitch_coef = new float[6];
Filters.bw_lpc(gamma1, interp_lpc, bw_lpc1, lpcSize);
Filters.bw_lpc(gamma2, interp_lpc, bw_lpc2, lpcSize);
Filters.filter_mem2(frmBuf, frmIdx, bw_lpc1, bw_lpc2, swBuf, swIdx,
frameSize, lpcSize, mem_sw_whole, 0);
Ltp.open_loop_nbest_pitch(swBuf, swIdx, min_pitch, max_pitch, frameSize,
nol_pitch, nol_pitch_coef, 6);
ol_pitch = nol_pitch[0];
ol_pitch_coef = nol_pitch_coef[0];
/*Try to remove pitch multiples*/
for (i = 1; i < 6; i++)
{
if ((nol_pitch_coef[i] > .85 * ol_pitch_coef) &&
(Math.Abs(nol_pitch[i] - ol_pitch / 2.0) <= 1 ||
Math.Abs(nol_pitch[i] - ol_pitch / 3.0) <= 1 ||
Math.Abs(nol_pitch[i] - ol_pitch / 4.0) <= 1 ||
Math.Abs(nol_pitch[i] - ol_pitch / 5.0) <= 1))
{
/*ol_pitch_coef=nol_pitch_coef[i];*/
ol_pitch = nol_pitch[i];
}
}
/*if (ol_pitch>50)
* ol_pitch/=2;*/
/*ol_pitch_coef = sqrt(ol_pitch_coef);*/
}
else
{
ol_pitch = 0;
ol_pitch_coef = 0;
}
/*Compute "real" excitation*/
Filters.fir_mem2(frmBuf, frmIdx, interp_lpc, excBuf, excIdx, frameSize, lpcSize, mem_exc);
/* Compute open-loop excitation gain */
ol_gain = 0;
for (i = 0; i < frameSize; i++)
{
ol_gain += excBuf[excIdx + i] * excBuf[excIdx + i];
}
ol_gain = (float)Math.Sqrt(1 + ol_gain / frameSize);
}
/*VBR stuff*/
if (vbr != null && (vbr_enabled != 0 || vad_enabled != 0))
{
if (abr_enabled != 0)
{
float qual_change = 0;
if (abr_drift2 * abr_drift > 0)
{
/* Only adapt if long-term and short-term drift are the same sign */
qual_change = -.00001f * abr_drift / (1 + abr_count);
if (qual_change > .05f)
{
qual_change = .05f;
}
if (qual_change < -.05f)
{
qual_change = -.05f;
}
}
vbr_quality += qual_change;
if (vbr_quality > 10)
{
vbr_quality = 10;
}
if (vbr_quality < 0)
{
vbr_quality = 0;
}
}
relative_quality = vbr.analysis(vin, frameSize, ol_pitch, ol_pitch_coef);
/*if (delta_qual<0)*/
/* delta_qual*=.1*(3+st->vbr_quality);*/
if (vbr_enabled != 0)
{
int mode;
int choice = 0;
float min_diff = 100;
mode = 8;
while (mode > 0)
{
int v1;
float thresh;
v1 = (int)Math.Floor(vbr_quality);
if (v1 == 10)
{
thresh = NSpeex.Vbr.nb_thresh[mode][v1];
}
else
{
thresh = (vbr_quality - v1) * NSpeex.Vbr.nb_thresh[mode][v1 + 1] +
(1 + v1 - vbr_quality) * NSpeex.Vbr.nb_thresh[mode][v1];
}
if (relative_quality > thresh &&
relative_quality - thresh < min_diff)
{
choice = mode;
min_diff = relative_quality - thresh;
}
mode--;
}
mode = choice;
if (mode == 0)
{
if (dtx_count == 0 || lsp_dist > .05 || dtx_enabled == 0 || dtx_count > 20)
{
mode = 1;
dtx_count = 1;
}
else
{
mode = 0;
dtx_count++;
}
}
else
{
dtx_count = 0;
}
Mode = mode;
if (abr_enabled != 0)
{
int bitrate;
bitrate = BitRate;
abr_drift += (bitrate - abr_enabled);
abr_drift2 = .95f * abr_drift2 + .05f * (bitrate - abr_enabled);
abr_count += 1.0f;
}
}
else
{
/*VAD only case*/
int mode;
if (relative_quality < 2)
{
if (dtx_count == 0 || lsp_dist > .05 || dtx_enabled == 0 || dtx_count > 20)
{
dtx_count = 1;
mode = 1;
}
else
{
mode = 0;
dtx_count++;
}
}
else
{
dtx_count = 0;
mode = submodeSelect;
}
/*speex_encoder_ctl(state, SPEEX_SET_MODE, &mode);*/
submodeID = mode;
}
}
else
{
relative_quality = -1;
}
/* First, transmit a zero for narrowband */
bits.Pack(0, 1);
/* Transmit the sub-mode we use for this frame */
bits.Pack(submodeID, NB_SUBMODE_BITS);
/* If null mode (no transmission), just set a couple things to zero*/
if (submodes[submodeID] == null)
{
for (i = 0; i < frameSize; i++)
{
excBuf[excIdx + i] = exc2Buf[exc2Idx + i] = swBuf[swIdx + i] = VERY_SMALL;
}
for (i = 0; i < lpcSize; i++)
{
mem_sw[i] = 0;
}
first = 1;
bounded_pitch = 1;
/* Final signal synthesis from excitation */
Filters.iir_mem2(excBuf, excIdx, interp_qlpc, frmBuf, frmIdx, frameSize, lpcSize, mem_sp);
vin[0] = frmBuf[frmIdx] + preemph * pre_mem2;
for (i = 1; i < frameSize; i++)
{
vin[i] = frmBuf[frmIdx = i] + preemph * vin[i - 1];
}
pre_mem2 = vin[frameSize - 1];
return(0);
}
/* LSP Quantization */
if (first != 0)
{
for (i = 0; i < lpcSize; i++)
{
old_lsp[i] = lsp[i];
}
}
/*Quantize LSPs*/
//#if 1 /*0 for unquantized*/
submodes[submodeID].lsqQuant.quant(lsp, qlsp, lpcSize, bits);
//#else
// for (i=0;i<lpcSize;i++)
// qlsp[i]=lsp[i];
//#endif
/*If we use low bit-rate pitch mode, transmit open-loop pitch*/
if (submodes[submodeID].lbr_pitch != -1)
{
bits.Pack(ol_pitch - min_pitch, 7);
}
if (submodes[submodeID].forced_pitch_gain != 0)
{
int quant;
quant = (int)Math.Floor(.5 + 15 * ol_pitch_coef);
if (quant > 15)
{
quant = 15;
}
if (quant < 0)
{
quant = 0;
}
bits.Pack(quant, 4);
ol_pitch_coef = (float)0.066667 * quant;
}
/*Quantize and transmit open-loop excitation gain*/
{
int qe = (int)(Math.Floor(0.5 + 3.5 * Math.Log(ol_gain)));
if (qe < 0)
{
qe = 0;
}
if (qe > 31)
{
qe = 31;
}
ol_gain = (float)Math.Exp(qe / 3.5);
bits.Pack(qe, 5);
}
/* Special case for first frame */
if (first != 0)
{
for (i = 0; i < lpcSize; i++)
{
old_qlsp[i] = qlsp[i];
}
}
/* Filter response */
res = new float[subframeSize];
/* Target signal */
target = new float[subframeSize];
syn_resp = new float[subframeSize];
mem = new float[lpcSize];
orig = new float[frameSize];
for (i = 0; i < frameSize; i++)
{
orig[i] = frmBuf[frmIdx + i];
}
/* Loop on sub-frames */
for (int sub = 0; sub < nbSubframes; sub++)
{
float tmp;
int offset;
int sp, sw, exc, exc2;
int pitchval;
/* Offset relative to start of frame */
offset = subframeSize * sub;
/* Original signal */
sp = frmIdx + offset;
/* Excitation */
exc = excIdx + offset;
/* Weighted signal */
sw = swIdx + offset;
exc2 = exc2Idx + offset;
/* LSP interpolation (quantized and unquantized) */
tmp = (float)(1.0 + sub) / nbSubframes;
for (i = 0; i < lpcSize; i++)
{
interp_lsp[i] = (1 - tmp) * old_lsp[i] + tmp * lsp[i];
}
for (i = 0; i < lpcSize; i++)
{
interp_qlsp[i] = (1 - tmp) * old_qlsp[i] + tmp * qlsp[i];
}
/* Make sure the filters are stable */
Lsp.enforce_margin(interp_lsp, lpcSize, .002f);
Lsp.enforce_margin(interp_qlsp, lpcSize, .002f);
/* Compute interpolated LPCs (quantized and unquantized) */
for (i = 0; i < lpcSize; i++)
{
interp_lsp[i] = (float)Math.Cos(interp_lsp[i]);
}
m_lsp.lsp2lpc(interp_lsp, interp_lpc, lpcSize);
for (i = 0; i < lpcSize; i++)
{
interp_qlsp[i] = (float)Math.Cos(interp_qlsp[i]);
}
m_lsp.lsp2lpc(interp_qlsp, interp_qlpc, lpcSize);
/* Compute analysis filter gain at w=pi (for use in SB-CELP) */
tmp = 1;
pi_gain[sub] = 0;
for (i = 0; i <= lpcSize; i++)
{
pi_gain[sub] += tmp * interp_qlpc[i];
tmp = -tmp;
}
/* Compute bandwidth-expanded (unquantized) LPCs for perceptual weighting */
Filters.bw_lpc(gamma1, interp_lpc, bw_lpc1, lpcSize);
if (gamma2 >= 0)
{
Filters.bw_lpc(gamma2, interp_lpc, bw_lpc2, lpcSize);
}
else
{
bw_lpc2[0] = 1;
bw_lpc2[1] = -preemph;
for (i = 2; i <= lpcSize; i++)
{
bw_lpc2[i] = 0;
}
}
/* Compute impulse response of A(z/g1) / ( A(z)*A(z/g2) )*/
for (i = 0; i < subframeSize; i++)
{
excBuf[exc + i] = 0;
}
excBuf[exc] = 1;
Filters.syn_percep_zero(excBuf, exc, interp_qlpc, bw_lpc1, bw_lpc2, syn_resp, subframeSize, lpcSize);
/* Reset excitation */
for (i = 0; i < subframeSize; i++)
{
excBuf[exc + i] = 0;
}
for (i = 0; i < subframeSize; i++)
{
exc2Buf[exc2 + i] = 0;
}
/* Compute zero response of A(z/g1) / ( A(z/g2) * A(z) ) */
for (i = 0; i < lpcSize; i++)
{
mem[i] = mem_sp[i];
}
Filters.iir_mem2(excBuf, exc, interp_qlpc, excBuf, exc, subframeSize, lpcSize, mem);
for (i = 0; i < lpcSize; i++)
{
mem[i] = mem_sw[i];
}
Filters.filter_mem2(excBuf, exc, bw_lpc1, bw_lpc2, res, 0, subframeSize, lpcSize, mem, 0);
/* Compute weighted signal */
for (i = 0; i < lpcSize; i++)
{
mem[i] = mem_sw[i];
}
Filters.filter_mem2(frmBuf, sp, bw_lpc1, bw_lpc2, swBuf, sw, subframeSize, lpcSize, mem, 0);
/* Compute target signal */
for (i = 0; i < subframeSize; i++)
{
target[i] = swBuf[sw + i] - res[i];
}
for (i = 0; i < subframeSize; i++)
{
excBuf[exc + i] = exc2Buf[exc2 + i] = 0;
}
/* If we have a long-term predictor (otherwise, something's wrong) */
// if (submodes[submodeID].ltp.quant)
// {
int pit_min, pit_max;
/* Long-term prediction */
if (submodes[submodeID].lbr_pitch != -1)
{
/* Low bit-rate pitch handling */
int margin;
margin = submodes[submodeID].lbr_pitch;
if (margin != 0)
{
if (ol_pitch < min_pitch + margin - 1)
{
ol_pitch = min_pitch + margin - 1;
}
if (ol_pitch > max_pitch - margin)
{
ol_pitch = max_pitch - margin;
}
pit_min = ol_pitch - margin + 1;
pit_max = ol_pitch + margin;
}
else
{
pit_min = pit_max = ol_pitch;
}
}
else
{
pit_min = min_pitch;
pit_max = max_pitch;
}
/* Force pitch to use only the current frame if needed */
if (bounded_pitch != 0 && pit_max > offset)
{
pit_max = offset;
}
/* Perform pitch search */
pitchval = submodes[submodeID].ltp.Quant(target, swBuf, sw, interp_qlpc, bw_lpc1, bw_lpc2,
excBuf, exc, pit_min, pit_max, ol_pitch_coef, lpcSize,
subframeSize, bits, exc2Buf, exc2, syn_resp, complexity);
pitch[sub] = pitchval;
// } else {
// speex_error ("No pitch prediction, what's wrong");
// }
/* Update target for adaptive codebook contribution */
Filters.syn_percep_zero(excBuf, exc, interp_qlpc, bw_lpc1, bw_lpc2, res, subframeSize, lpcSize);
for (i = 0; i < subframeSize; i++)
{
target[i] -= res[i];
}
/* Quantization of innovation */
{
int innovptr;
float ener = 0, ener_1;
innovptr = sub * subframeSize;
for (i = 0; i < subframeSize; i++)
{
innov[innovptr + i] = 0;
}
Filters.residue_percep_zero(target, 0, interp_qlpc, bw_lpc1, bw_lpc2, buf2, subframeSize, lpcSize);
for (i = 0; i < subframeSize; i++)
{
ener += buf2[i] * buf2[i];
}
ener = (float)Math.Sqrt(.1f + ener / subframeSize);
/*for (i=0;i<subframeSize;i++)
* System.out.print(buf2[i]/ener + "\t");
*/
ener /= ol_gain;
/* Calculate gain correction for the sub-frame (if any) */
if (submodes[submodeID].have_subframe_gain != 0)
{
int qe;
ener = (float)Math.Log(ener);
if (submodes[submodeID].have_subframe_gain == 3)
{
qe = VQ.index(ener, exc_gain_quant_scal3, 8);
bits.Pack(qe, 3);
ener = exc_gain_quant_scal3[qe];
}
else
{
qe = VQ.index(ener, exc_gain_quant_scal1, 2);
bits.Pack(qe, 1);
ener = exc_gain_quant_scal1[qe];
}
ener = (float)Math.Exp(ener);
}
else
{
ener = 1;
}
ener *= ol_gain;
/*System.out.println(ener + " " + ol_gain);*/
ener_1 = 1 / ener;
/* Normalize innovation */
for (i = 0; i < subframeSize; i++)
{
target[i] *= ener_1;
}
/* Quantize innovation */
// if (submodes[submodeID].innovation != null)
// {
/* Codebook search */
submodes[submodeID].innovation.Quant(target, interp_qlpc, bw_lpc1, bw_lpc2,
lpcSize, subframeSize, innov,
innovptr, syn_resp, bits, complexity);
/* De-normalize innovation and update excitation */
for (i = 0; i < subframeSize; i++)
{
innov[innovptr + i] *= ener;
}
for (i = 0; i < subframeSize; i++)
{
excBuf[exc + i] += innov[innovptr + i];
}
// } else {
// speex_error("No fixed codebook");
// }
/* In some (rare) modes, we do a second search (more bits) to reduce noise even more */
if (submodes[submodeID].double_codebook != 0)
{
float[] innov2 = new float[subframeSize];
// for (i=0;i<subframeSize;i++)
// innov2[i]=0;
for (i = 0; i < subframeSize; i++)
{
target[i] *= 2.2f;
}
submodes[submodeID].innovation.Quant(target, interp_qlpc, bw_lpc1, bw_lpc2,
lpcSize, subframeSize, innov2, 0,
syn_resp, bits, complexity);
for (i = 0; i < subframeSize; i++)
{
innov2[i] *= ener * (1f / 2.2f);
}
for (i = 0; i < subframeSize; i++)
{
excBuf[exc + i] += innov2[i];
}
}
for (i = 0; i < subframeSize; i++)
{
target[i] *= ener;
}
}
/*Keep the previous memory*/
for (i = 0; i < lpcSize; i++)
{
mem[i] = mem_sp[i];
}
/* Final signal synthesis from excitation */
Filters.iir_mem2(excBuf, exc, interp_qlpc, frmBuf, sp, subframeSize, lpcSize, mem_sp);
/* Compute weighted signal again, from synthesized speech (not sure it's the right thing) */
Filters.filter_mem2(frmBuf, sp, bw_lpc1, bw_lpc2, swBuf, sw, subframeSize, lpcSize, mem_sw, 0);
for (i = 0; i < subframeSize; i++)
{
exc2Buf[exc2 + i] = excBuf[exc + i];
}
}
/* Store the LSPs for interpolation in the next frame */
if (submodeID >= 1)
{
for (i = 0; i < lpcSize; i++)
{
old_lsp[i] = lsp[i];
}
for (i = 0; i < lpcSize; i++)
{
old_qlsp[i] = qlsp[i];
}
}
if (submodeID == 1)
{
if (dtx_count != 0)
{
bits.Pack(15, 4);
}
else
{
bits.Pack(0, 4);
}
}
/* The next frame will not be the first (Duh!) */
first = 0;
{
float ener = 0, err = 0;
float snr;
for (i = 0; i < frameSize; i++)
{
ener += frmBuf[frmIdx + i] * frmBuf[frmIdx + i];
err += (frmBuf[frmIdx + i] - orig[i]) * (frmBuf[frmIdx + i] - orig[i]);
}
snr = (float)(10 * Math.Log((ener + 1) / (err + 1)));
/*System.out.println("Frame result: SNR="+snr+" E="+ener+" Err="+err+"\r\n");*/
}
/* Replace input by synthesized speech */
vin[0] = frmBuf[frmIdx] + preemph * pre_mem2;
for (i = 1; i < frameSize; i++)
{
vin[i] = frmBuf[frmIdx + i] + preemph * vin[i - 1];
}
pre_mem2 = vin[frameSize - 1];
if (submodes[submodeID].innovation is NoiseSearch || submodeID == 0)
{
bounded_pitch = 1;
}
else
{
bounded_pitch = 0;
}
return(1);
}
/// <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);
}