internal static void unquant_coarse_energy(CeltMode m, int start, int end, int[] oldEBands, int intra, EntropyCoder dec, int C, int LM)
{
byte[] prob_model = Tables.e_prob_model[LM][intra];
int i, c;
int[] prev = { 0, 0 };
int coef;
int beta;
int budget;
int tell;
if (intra != 0)
{
coef = 0;
beta = beta_intra;
}
else
{
beta = beta_coef[LM];
coef = pred_coef[LM];
}
budget = (int)dec.storage * 8;
/* Decode at a fixed coarse resolution */
for (i = start; i < end; i++)
{
c = 0;
do
{
int qi;
int q;
int tmp;
/* It would be better to express this invariant as a
* test on C at function entry, but that isn't enough
* to make the static analyzer happy. */
Inlines.OpusAssert(c < 2);
tell = dec.tell();
if (budget - tell >= 15)
{
int pi;
pi = 2 * Inlines.IMIN(i, 20);
qi = Laplace.ec_laplace_decode(dec,
(uint)prob_model[pi] << 7, prob_model[pi + 1] << 6);
}
else if (budget - tell >= 2)
{
qi = dec.dec_icdf(small_energy_icdf, 2);
qi = (qi >> 1) ^ -(qi & 1);
}
else if (budget - tell >= 1)
{
qi = 0 - dec.dec_bit_logp(1);
}
else
{
qi = -1;
}
q = (int)Inlines.SHL32(qi, CeltConstants.DB_SHIFT); // opus bug: useless extend32
oldEBands[i + c * m.nbEBands] = Inlines.MAX16((0 - ((short)(0.5 + (9.0f) * (((int)1) << (CeltConstants.DB_SHIFT)))) /*Inlines.QCONST16(9.0f, CeltConstants.DB_SHIFT)*/), oldEBands[i + c * m.nbEBands]);
tmp = Inlines.PSHR32(Inlines.MULT16_16(coef, oldEBands[i + c * m.nbEBands]), 8) + prev[c] + Inlines.SHL32(q, 7);
tmp = Inlines.MAX32(-((int)(0.5 + (28.0f) * (((int)1) << (CeltConstants.DB_SHIFT + 7)))) /*Inlines.QCONST32(28.0f, CeltConstants.DB_SHIFT + 7)*/, tmp);
oldEBands[i + c * m.nbEBands] = (Inlines.PSHR32(tmp, 7));
prev[c] = prev[c] + Inlines.SHL32(q, 7) - Inlines.MULT16_16(beta, Inlines.PSHR32(q, 8));
} while (++c < C);
}
}
internal int opus_decode_frame(byte[] data, int data_ptr,
int len, short[] pcm, int pcm_ptr, int frame_size, int decode_fec)
{
SilkDecoder silk_dec;
CeltDecoder celt_dec;
int i, silk_ret = 0, celt_ret = 0;
EntropyCoder dec = new EntropyCoder(); // porting note: stack var
int silk_frame_size;
int pcm_silk_size;
short[] pcm_silk;
int pcm_transition_silk_size;
short[] pcm_transition_silk;
int pcm_transition_celt_size;
short[] pcm_transition_celt;
short[] pcm_transition = null;
int redundant_audio_size;
short[] redundant_audio;
int audiosize;
OpusMode mode;
int transition = 0;
int start_band;
int redundancy = 0;
int redundancy_bytes = 0;
int celt_to_silk = 0;
int c;
int F2_5, F5, F10, F20;
int[] window;
uint redundant_rng = 0;
int celt_accum;
silk_dec = this.SilkDecoder;
celt_dec = this.Celt_Decoder;
F20 = this.Fs / 50;
F10 = F20 >> 1;
F5 = F10 >> 1;
F2_5 = F5 >> 1;
if (frame_size < F2_5)
{
return(OpusError.OPUS_BUFFER_TOO_SMALL);
}
/* Limit frame_size to avoid excessive stack allocations. */
frame_size = Inlines.IMIN(frame_size, this.Fs / 25 * 3);
/* Payloads of 1 (2 including ToC) or 0 trigger the PLC/DTX */
if (len <= 1)
{
data = null;
/* In that case, don't conceal more than what the ToC says */
frame_size = Inlines.IMIN(frame_size, this.frame_size);
}
if (data != null)
{
audiosize = this.frame_size;
mode = this.mode;
dec.dec_init(data, data_ptr, (uint)len);
}
else
{
audiosize = frame_size;
mode = this.prev_mode;
if (mode == 0)
{
/* If we haven't got any packet yet, all we can do is return zeros */
for (i = pcm_ptr; i < pcm_ptr + (audiosize * this.channels); i++)
{
pcm[i] = 0;
}
return(audiosize);
}
/* Avoids trying to run the PLC on sizes other than 2.5 (CELT), 5 (CELT),
* 10, or 20 (e.g. 12.5 or 30 ms). */
if (audiosize > F20)
{
do
{
int ret = opus_decode_frame(null, 0, 0, pcm, pcm_ptr, Inlines.IMIN(audiosize, F20), 0);
if (ret < 0)
{
return(ret);
}
pcm_ptr += ret * this.channels;
audiosize -= ret;
} while (audiosize > 0);
return(frame_size);
}
else if (audiosize < F20)
{
if (audiosize > F10)
{
audiosize = F10;
}
else if (mode != OpusMode.MODE_SILK_ONLY && audiosize > F5 && audiosize < F10)
{
audiosize = F5;
}
}
}
/* In fixed-point, we can tell CELT to do the accumulation on top of the
* SILK PCM buffer. This saves some stack space. */
celt_accum = ((mode != OpusMode.MODE_CELT_ONLY) && (frame_size >= F10)) ? 1 : 0;
pcm_transition_silk_size = 0;
pcm_transition_celt_size = 0;
if (data != null && this.prev_mode > 0 && (
(mode == OpusMode.MODE_CELT_ONLY && this.prev_mode != OpusMode.MODE_CELT_ONLY && (this.prev_redundancy == 0)) ||
(mode != OpusMode.MODE_CELT_ONLY && this.prev_mode == OpusMode.MODE_CELT_ONLY))
)
{
transition = 1;
/* Decide where to allocate the stack memory for pcm_transition */
if (mode == OpusMode.MODE_CELT_ONLY)
{
pcm_transition_celt_size = F5 * this.channels;
}
else
{
pcm_transition_silk_size = F5 * this.channels;
}
}
pcm_transition_celt = new short[pcm_transition_celt_size];
if (transition != 0 && mode == OpusMode.MODE_CELT_ONLY)
{
pcm_transition = pcm_transition_celt;
opus_decode_frame(null, 0, 0, pcm_transition, 0, Inlines.IMIN(F5, audiosize), 0);
}
if (audiosize > frame_size)
{
/*fprintf(stderr, "PCM buffer too small: %d vs %d (mode = %d)\n", audiosize, frame_size, mode);*/
return(OpusError.OPUS_BAD_ARG);
}
else
{
frame_size = audiosize;
}
/* Don't allocate any memory when in CELT-only mode */
pcm_silk_size = (mode != OpusMode.MODE_CELT_ONLY && (celt_accum == 0)) ? Inlines.IMAX(F10, frame_size) * this.channels : 0;
pcm_silk = new short[pcm_silk_size];
/* SILK processing */
if (mode != OpusMode.MODE_CELT_ONLY)
{
int lost_flag, decoded_samples;
short[] pcm_ptr2;
int pcm_ptr2_ptr = 0;
if (celt_accum != 0)
{
pcm_ptr2 = pcm;
pcm_ptr2_ptr = pcm_ptr;
}
else
{
pcm_ptr2 = pcm_silk;
pcm_ptr2_ptr = 0;
}
if (this.prev_mode == OpusMode.MODE_CELT_ONLY)
{
DecodeAPI.silk_InitDecoder(silk_dec);
}
/* The SILK PLC cannot produce frames of less than 10 ms */
this.DecControl.payloadSize_ms = Inlines.IMAX(10, 1000 * audiosize / this.Fs);
if (data != null)
{
this.DecControl.nChannelsInternal = this.stream_channels;
if (mode == OpusMode.MODE_SILK_ONLY)
{
if (this.bandwidth == OpusBandwidth.OPUS_BANDWIDTH_NARROWBAND)
{
this.DecControl.internalSampleRate = 8000;
}
else if (this.bandwidth == OpusBandwidth.OPUS_BANDWIDTH_MEDIUMBAND)
{
this.DecControl.internalSampleRate = 12000;
}
else if (this.bandwidth == OpusBandwidth.OPUS_BANDWIDTH_WIDEBAND)
{
this.DecControl.internalSampleRate = 16000;
}
else
{
this.DecControl.internalSampleRate = 16000;
Inlines.OpusAssert(false);
}
}
else
{
/* Hybrid mode */
this.DecControl.internalSampleRate = 16000;
}
}
lost_flag = data == null ? 1 : 2 * decode_fec;
decoded_samples = 0;
do
{
/* Call SILK decoder */
int first_frame = (decoded_samples == 0) ? 1 : 0;
silk_ret = DecodeAPI.silk_Decode(silk_dec, this.DecControl,
lost_flag, first_frame, dec, pcm_ptr2, pcm_ptr2_ptr, out silk_frame_size);
if (silk_ret != 0)
{
if (lost_flag != 0)
{
/* PLC failure should not be fatal */
silk_frame_size = frame_size;
Arrays.MemSetWithOffset <short>(pcm_ptr2, 0, pcm_ptr2_ptr, frame_size * this.channels);
}
else
{
return(OpusError.OPUS_INTERNAL_ERROR);
}
}
pcm_ptr2_ptr += (silk_frame_size * this.channels);
decoded_samples += silk_frame_size;
} while (decoded_samples < frame_size);
}
start_band = 0;
if (decode_fec == 0 && mode != OpusMode.MODE_CELT_ONLY && data != null &&
dec.tell() + 17 + 20 * (this.mode == OpusMode.MODE_HYBRID ? 1 : 0) <= 8 * len)
{
/* Check if we have a redundant 0-8 kHz band */
if (mode == OpusMode.MODE_HYBRID)
{
redundancy = dec.dec_bit_logp(12);
}
else
{
redundancy = 1;
}
if (redundancy != 0)
{
celt_to_silk = dec.dec_bit_logp(1);
/* redundancy_bytes will be at least two, in the non-hybrid
* case due to the ec_tell() check above */
redundancy_bytes = mode == OpusMode.MODE_HYBRID ?
(int)dec.dec_uint(256) + 2 :
len - ((dec.tell() + 7) >> 3);
len -= redundancy_bytes;
/* This is a sanity check. It should never happen for a valid
* packet, so the exact behaviour is not normative. */
if (len * 8 < dec.tell())
{
len = 0;
redundancy_bytes = 0;
redundancy = 0;
}
/* Shrink decoder because of raw bits */
dec.storage = (uint)(dec.storage - redundancy_bytes);
}
}
if (mode != OpusMode.MODE_CELT_ONLY)
{
start_band = 17;
}
{
int endband = 21;
switch (this.bandwidth)
{
case OpusBandwidth.OPUS_BANDWIDTH_NARROWBAND:
endband = 13;
break;
case OpusBandwidth.OPUS_BANDWIDTH_MEDIUMBAND:
case OpusBandwidth.OPUS_BANDWIDTH_WIDEBAND:
endband = 17;
break;
case OpusBandwidth.OPUS_BANDWIDTH_SUPERWIDEBAND:
endband = 19;
break;
case OpusBandwidth.OPUS_BANDWIDTH_FULLBAND:
endband = 21;
break;
}
celt_dec.SetEndBand(endband);
celt_dec.SetChannels(this.stream_channels);
}
if (redundancy != 0)
{
transition = 0;
pcm_transition_silk_size = 0;
}
pcm_transition_silk = new short[pcm_transition_silk_size];
if (transition != 0 && mode != OpusMode.MODE_CELT_ONLY)
{
pcm_transition = pcm_transition_silk;
opus_decode_frame(null, 0, 0, pcm_transition, 0, Inlines.IMIN(F5, audiosize), 0);
}
/* Only allocation memory for redundancy if/when needed */
redundant_audio_size = redundancy != 0 ? F5 * this.channels : 0;
redundant_audio = new short[redundant_audio_size];
/* 5 ms redundant frame for CELT->SILK*/
if (redundancy != 0 && celt_to_silk != 0)
{
celt_dec.SetStartBand(0);
celt_dec.celt_decode_with_ec(data, (data_ptr + len), redundancy_bytes,
redundant_audio, 0, F5, null, 0);
redundant_rng = celt_dec.GetFinalRange();
}
/* MUST be after PLC */
celt_dec.SetStartBand(start_band);
if (mode != OpusMode.MODE_SILK_ONLY)
{
int celt_frame_size = Inlines.IMIN(F20, frame_size);
/* Make sure to discard any previous CELT state */
if (mode != this.prev_mode && this.prev_mode > 0 && this.prev_redundancy == 0)
{
celt_dec.ResetState();
}
/* Decode CELT */
celt_ret = celt_dec.celt_decode_with_ec(decode_fec != 0 ? null : data, data_ptr,
len, pcm, pcm_ptr, celt_frame_size, dec, celt_accum);
}
else
{
if (celt_accum == 0)
{
for (i = pcm_ptr; i < (frame_size * this.channels) + pcm_ptr; i++)
{
pcm[i] = 0;
}
}
/* For hybrid -> SILK transitions, we let the CELT MDCT
* do a fade-out by decoding a silence frame */
if (this.prev_mode == OpusMode.MODE_HYBRID && !(redundancy != 0 && celt_to_silk != 0 && this.prev_redundancy != 0))
{
celt_dec.SetStartBand(0);
celt_dec.celt_decode_with_ec(SILENCE, 0, 2, pcm, pcm_ptr, F2_5, null, celt_accum);
}
}
if (mode != OpusMode.MODE_CELT_ONLY && celt_accum == 0)
{
for (i = 0; i < frame_size * this.channels; i++)
{
pcm[pcm_ptr + i] = Inlines.SAT16(Inlines.ADD32(pcm[pcm_ptr + i], pcm_silk[i]));
}
}
window = celt_dec.GetMode().window;
/* 5 ms redundant frame for SILK->CELT */
if (redundancy != 0 && celt_to_silk == 0)
{
celt_dec.ResetState();
celt_dec.SetStartBand(0);
celt_dec.celt_decode_with_ec(data, data_ptr + len, redundancy_bytes, redundant_audio, 0, F5, null, 0);
redundant_rng = celt_dec.GetFinalRange();
CodecHelpers.smooth_fade(pcm, pcm_ptr + this.channels * (frame_size - F2_5), redundant_audio, this.channels * F2_5,
pcm, (pcm_ptr + this.channels * (frame_size - F2_5)), F2_5, this.channels, window, this.Fs);
}
if (redundancy != 0 && celt_to_silk != 0)
{
for (c = 0; c < this.channels; c++)
{
for (i = 0; i < F2_5; i++)
{
pcm[this.channels * i + c + pcm_ptr] = redundant_audio[this.channels * i + c];
}
}
CodecHelpers.smooth_fade(redundant_audio, (this.channels * F2_5), pcm, (pcm_ptr + (this.channels * F2_5)),
pcm, (pcm_ptr + (this.channels * F2_5)), F2_5, this.channels, window, this.Fs);
}
if (transition != 0)
{
if (audiosize >= F5)
{
for (i = 0; i < this.channels * F2_5; i++)
{
pcm[i] = pcm_transition[i];
}
CodecHelpers.smooth_fade(pcm_transition, (this.channels * F2_5), pcm, (pcm_ptr + (this.channels * F2_5)),
pcm, (pcm_ptr + (this.channels * F2_5)), F2_5,
this.channels, window, this.Fs);
}
else
{
/* Not enough time to do a clean transition, but we do it anyway
* This will not preserve amplitude perfectly and may introduce
* a bit of temporal aliasing, but it shouldn't be too bad and
* that's pretty much the best we can do. In any case, generating this
* transition is pretty silly in the first place */
CodecHelpers.smooth_fade(pcm_transition, 0, pcm, pcm_ptr,
pcm, pcm_ptr, F2_5,
this.channels, window, this.Fs);
}
}
if (this.decode_gain != 0)
{
int gain;
gain = Inlines.celt_exp2(Inlines.MULT16_16_P15(((short)(0.5 + (6.48814081e-4f) * (((int)1) << (25)))) /*Inlines.QCONST16(6.48814081e-4f, 25)*/, this.decode_gain));
for (i = pcm_ptr; i < pcm_ptr + (frame_size * this.channels); i++)
{
int x;
x = Inlines.MULT16_32_P16(pcm[i], gain);
pcm[i] = (short)Inlines.SATURATE(x, 32767);
}
}
if (len <= 1)
{
this.rangeFinal = 0;
}
else
{
this.rangeFinal = dec.rng ^ redundant_rng;
}
this.prev_mode = mode;
this.prev_redundancy = (redundancy != 0 && celt_to_silk == 0) ? 1 : 0;
return(celt_ret < 0 ? celt_ret : audiosize);
}
internal static int interp_bits2pulses(CeltMode m, int start, int end, int skip_start,
int[] bits1, int[] bits2, int[] thresh, int[] cap, int total, out int _balance,
int skip_rsv, ref int intensity, int intensity_rsv, ref int dual_stereo, int dual_stereo_rsv, int[] bits,
int[] ebits, int[] fine_priority, int C, int LM, EntropyCoder ec, int encode, int prev, int signalBandwidth)
{
int psum;
int lo, hi;
int i, j;
int logM;
int stereo;
int codedBands = -1;
int alloc_floor;
int left, percoeff;
int done;
int balance;
alloc_floor = C << EntropyCoder.BITRES;
stereo = C > 1 ? 1 : 0;
logM = LM << EntropyCoder.BITRES;
lo = 0;
hi = 1 << ALLOC_STEPS;
for (i = 0; i < ALLOC_STEPS; i++)
{
int mid = (lo + hi) >> 1;
psum = 0;
done = 0;
for (j = end; j-- > start;)
{
int tmp = bits1[j] + (mid * (int)bits2[j] >> ALLOC_STEPS);
if (tmp >= thresh[j] || done != 0)
{
done = 1;
/* Don't allocate more than we can actually use */
psum += Inlines.IMIN(tmp, cap[j]);
}
else
{
if (tmp >= alloc_floor)
{
psum += alloc_floor;
}
}
}
if (psum > total)
{
hi = mid;
}
else
{
lo = mid;
}
}
psum = 0;
/*printf ("interp bisection gave %d\n", lo);*/
done = 0;
for (j = end; j-- > start;)
{
int tmp = bits1[j] + (lo * bits2[j] >> ALLOC_STEPS);
if (tmp < thresh[j] && done == 0)
{
if (tmp >= alloc_floor)
{
tmp = alloc_floor;
}
else
{
tmp = 0;
}
}
else
{
done = 1;
}
/* Don't allocate more than we can actually use */
tmp = Inlines.IMIN(tmp, cap[j]);
bits[j] = tmp;
psum += tmp;
}
/* Decide which bands to skip, working backwards from the end. */
for (codedBands = end; ; codedBands--)
{
int band_width;
int band_bits;
int rem;
j = codedBands - 1;
/* Never skip the first band, nor a band that has been boosted by
* dynalloc.
* In the first case, we'd be coding a bit to signal we're going to waste
* all the other bits.
* In the second case, we'd be coding a bit to redistribute all the bits
* we just signaled should be cocentrated in this band. */
if (j <= skip_start)
{
/* Give the bit we reserved to end skipping back. */
total += skip_rsv;
break;
}
/*Figure out how many left-over bits we would be adding to this band.
* This can include bits we've stolen back from higher, skipped bands.*/
left = total - psum;
percoeff = Inlines.celt_udiv(left, m.eBands[codedBands] - m.eBands[start]);
left -= (m.eBands[codedBands] - m.eBands[start]) * percoeff;
rem = Inlines.IMAX(left - (m.eBands[j] - m.eBands[start]), 0);
band_width = m.eBands[codedBands] - m.eBands[j];
band_bits = (int)(bits[j] + percoeff * band_width + rem);
/*Only code a skip decision if we're above the threshold for this band.
* Otherwise it is force-skipped.
* This ensures that we have enough bits to code the skip flag.*/
if (band_bits >= Inlines.IMAX(thresh[j], alloc_floor + (1 << EntropyCoder.BITRES)))
{
if (encode != 0)
{
/*This if() block is the only part of the allocation function that
* is not a mandatory part of the bitstream: any bands we choose to
* skip here must be explicitly signaled.*/
/*Choose a threshold with some hysteresis to keep bands from
* fluctuating in and out.*/
#if FUZZING
if ((new Random().Next() & 0x1) == 0)
#else
if (codedBands <= start + 2 || (band_bits > ((j < prev ? 7 : 9) * band_width << LM << EntropyCoder.BITRES) >> 4 && j <= signalBandwidth))
#endif
{
ec.enc_bit_logp(1, 1);
break;
}
ec.enc_bit_logp(0, 1);
}
else if (ec.dec_bit_logp(1) != 0)
{
break;
}
/*We used a bit to skip this band.*/
psum += 1 << EntropyCoder.BITRES;
band_bits -= 1 << EntropyCoder.BITRES;
}
/*Reclaim the bits originally allocated to this band.*/
psum -= bits[j] + intensity_rsv;
if (intensity_rsv > 0)
{
intensity_rsv = LOG2_FRAC_TABLE[j - start];
}
psum += intensity_rsv;
if (band_bits >= alloc_floor)
{
/*If we have enough for a fine energy bit per channel, use it.*/
psum += alloc_floor;
bits[j] = alloc_floor;
}
else
{
/*Otherwise this band gets nothing at all.*/
bits[j] = 0;
}
}
Inlines.OpusAssert(codedBands > start);
/* Code the intensity and dual stereo parameters. */
if (intensity_rsv > 0)
{
if (encode != 0)
{
intensity = Inlines.IMIN(intensity, codedBands);
ec.enc_uint((uint)(intensity - start), (uint)(codedBands + 1 - start));
}
else
{
intensity = start + (int)ec.dec_uint((uint)(codedBands + 1 - start));
}
}
else
{
intensity = 0;
}
if (intensity <= start)
{
total += dual_stereo_rsv;
dual_stereo_rsv = 0;
}
if (dual_stereo_rsv > 0)
{
if (encode != 0)
{
ec.enc_bit_logp(dual_stereo, 1);
}
else
{
dual_stereo = ec.dec_bit_logp(1);
}
}
else
{
dual_stereo = 0;
}
/* Allocate the remaining bits */
left = total - psum;
percoeff = Inlines.celt_udiv(left, m.eBands[codedBands] - m.eBands[start]);
left -= (m.eBands[codedBands] - m.eBands[start]) * percoeff;
for (j = start; j < codedBands; j++)
{
bits[j] += ((int)percoeff * (m.eBands[j + 1] - m.eBands[j]));
}
for (j = start; j < codedBands; j++)
{
int tmp = (int)Inlines.IMIN(left, m.eBands[j + 1] - m.eBands[j]);
bits[j] += tmp;
left -= tmp;
}
/*for (j=0;j<end;j++)printf("%d ", bits[j]);printf("\n");*/
balance = 0;
for (j = start; j < codedBands; j++)
{
int N0, N, den;
int offset;
int NClogN;
int excess, bit;
Inlines.OpusAssert(bits[j] >= 0);
N0 = m.eBands[j + 1] - m.eBands[j];
N = N0 << LM;
bit = (int)bits[j] + balance;
if (N > 1)
{
excess = Inlines.MAX32(bit - cap[j], 0);
bits[j] = bit - excess;
/* Compensate for the extra DoF in stereo */
den = (C * N + ((C == 2 && N > 2 && (dual_stereo == 0) && j < intensity) ? 1 : 0));
NClogN = den * (m.logN[j] + logM);
/* Offset for the number of fine bits by log2(N)/2 + FINE_OFFSET
* compared to their "fair share" of total/N */
offset = (NClogN >> 1) - den * CeltConstants.FINE_OFFSET;
/* N=2 is the only point that doesn't match the curve */
if (N == 2)
{
offset += den << EntropyCoder.BITRES >> 2;
}
/* Changing the offset for allocating the second and third
* fine energy bit */
if (bits[j] + offset < den * 2 << EntropyCoder.BITRES)
{
offset += NClogN >> 2;
}
else if (bits[j] + offset < den * 3 << EntropyCoder.BITRES)
{
offset += NClogN >> 3;
}
/* Divide with rounding */
ebits[j] = Inlines.IMAX(0, (bits[j] + offset + (den << (EntropyCoder.BITRES - 1))));
ebits[j] = Inlines.celt_udiv(ebits[j], den) >> EntropyCoder.BITRES;
/* Make sure not to bust */
if (C * ebits[j] > (bits[j] >> EntropyCoder.BITRES))
{
ebits[j] = bits[j] >> stereo >> EntropyCoder.BITRES;
}
/* More than that is useless because that's about as far as PVQ can go */
ebits[j] = Inlines.IMIN(ebits[j], CeltConstants.MAX_FINE_BITS);
/* If we rounded down or capped this band, make it a candidate for the
* final fine energy pass */
fine_priority[j] = (ebits[j] * (den << EntropyCoder.BITRES) >= bits[j] + offset) ? 1 : 0;
/* Remove the allocated fine bits; the rest are assigned to PVQ */
bits[j] -= C * ebits[j] << EntropyCoder.BITRES;
}
else
{
/* For N=1, all bits go to fine energy except for a single sign bit */
excess = Inlines.MAX32(0, bit - (C << EntropyCoder.BITRES));
bits[j] = bit - excess;
ebits[j] = 0;
fine_priority[j] = 1;
}
/* Fine energy can't take advantage of the re-balancing in
* quant_all_bands().
* Instead, do the re-balancing here.*/
if (excess > 0)
{
int extra_fine;
int extra_bits;
extra_fine = Inlines.IMIN(excess >> (stereo + EntropyCoder.BITRES), CeltConstants.MAX_FINE_BITS - ebits[j]);
ebits[j] += extra_fine;
extra_bits = extra_fine * C << EntropyCoder.BITRES;
fine_priority[j] = (extra_bits >= excess - balance) ? 1 : 0;
excess -= extra_bits;
}
balance = excess;
Inlines.OpusAssert(bits[j] >= 0);
Inlines.OpusAssert(ebits[j] >= 0);
}
/* Save any remaining bits over the cap for the rebalancing in
* quant_all_bands(). */
_balance = balance;
/* The skipped bands use all their bits for fine energy. */
for (; j < end; j++)
{
ebits[j] = bits[j] >> stereo >> EntropyCoder.BITRES;
Inlines.OpusAssert(C * ebits[j] << EntropyCoder.BITRES == bits[j]);
bits[j] = 0;
fine_priority[j] = (ebits[j] < 1) ? 1 : 0;
}
return(codedBands);
}
/* Decode a frame */
internal static int silk_Decode( /* O Returns error code */
SilkDecoder psDec, /* I/O State */
DecControlState decControl, /* I/O Control Structure */
int lostFlag, /* I 0: no loss, 1 loss, 2 decode fec */
int newPacketFlag, /* I Indicates first decoder call for this packet */
EntropyCoder psRangeDec, /* I/O Compressor data structure */
short[] samplesOut, /* O Decoded output speech vector */
int samplesOut_ptr,
out int nSamplesOut /* O Number of samples decoded */
)
{
int i, n, decode_only_middle = 0, ret = SilkError.SILK_NO_ERROR;
int LBRR_symbol;
BoxedValueInt nSamplesOutDec = new BoxedValueInt();
short[] samplesOut_tmp;
int[] samplesOut_tmp_ptrs = new int[2];
short[] samplesOut1_tmp_storage1;
short[] samplesOut1_tmp_storage2;
short[] samplesOut2_tmp;
int[] MS_pred_Q13 = new int[] { 0, 0 };
short[] resample_out;
int resample_out_ptr;
SilkChannelDecoder[] channel_state = psDec.channel_state;
int has_side;
int stereo_to_mono;
int delay_stack_alloc;
nSamplesOut = 0;
Inlines.OpusAssert(decControl.nChannelsInternal == 1 || decControl.nChannelsInternal == 2);
/**********************************/
/* Test if first frame in payload */
/**********************************/
if (newPacketFlag != 0)
{
for (n = 0; n < decControl.nChannelsInternal; n++)
{
channel_state[n].nFramesDecoded = 0; /* Used to count frames in packet */
}
}
/* If Mono . Stereo transition in bitstream: init state of second channel */
if (decControl.nChannelsInternal > psDec.nChannelsInternal)
{
ret += channel_state[1].silk_init_decoder();
}
stereo_to_mono = (decControl.nChannelsInternal == 1 && psDec.nChannelsInternal == 2 &&
(decControl.internalSampleRate == 1000 * channel_state[0].fs_kHz)) ? 1 : 0;
if (channel_state[0].nFramesDecoded == 0)
{
for (n = 0; n < decControl.nChannelsInternal; n++)
{
int fs_kHz_dec;
if (decControl.payloadSize_ms == 0)
{
/* Assuming packet loss, use 10 ms */
channel_state[n].nFramesPerPacket = 1;
channel_state[n].nb_subfr = 2;
}
else if (decControl.payloadSize_ms == 10)
{
channel_state[n].nFramesPerPacket = 1;
channel_state[n].nb_subfr = 2;
}
else if (decControl.payloadSize_ms == 20)
{
channel_state[n].nFramesPerPacket = 1;
channel_state[n].nb_subfr = 4;
}
else if (decControl.payloadSize_ms == 40)
{
channel_state[n].nFramesPerPacket = 2;
channel_state[n].nb_subfr = 4;
}
else if (decControl.payloadSize_ms == 60)
{
channel_state[n].nFramesPerPacket = 3;
channel_state[n].nb_subfr = 4;
}
else
{
Inlines.OpusAssert(false);
return(SilkError.SILK_DEC_INVALID_FRAME_SIZE);
}
fs_kHz_dec = (decControl.internalSampleRate >> 10) + 1;
if (fs_kHz_dec != 8 && fs_kHz_dec != 12 && fs_kHz_dec != 16)
{
Inlines.OpusAssert(false);
return(SilkError.SILK_DEC_INVALID_SAMPLING_FREQUENCY);
}
ret += channel_state[n].silk_decoder_set_fs(fs_kHz_dec, decControl.API_sampleRate);
}
}
if (decControl.nChannelsAPI == 2 && decControl.nChannelsInternal == 2 && (psDec.nChannelsAPI == 1 || psDec.nChannelsInternal == 1))
{
Arrays.MemSetShort(psDec.sStereo.pred_prev_Q13, 0, 2);
Arrays.MemSetShort(psDec.sStereo.sSide, 0, 2);
channel_state[1].resampler_state.Assign(channel_state[0].resampler_state);
}
psDec.nChannelsAPI = decControl.nChannelsAPI;
psDec.nChannelsInternal = decControl.nChannelsInternal;
if (decControl.API_sampleRate > (int)SilkConstants.MAX_API_FS_KHZ * 1000 || decControl.API_sampleRate < 8000)
{
ret = SilkError.SILK_DEC_INVALID_SAMPLING_FREQUENCY;
return(ret);
}
if (lostFlag != DecoderAPIFlag.FLAG_PACKET_LOST && channel_state[0].nFramesDecoded == 0)
{
/* First decoder call for this payload */
/* Decode VAD flags and LBRR flag */
for (n = 0; n < decControl.nChannelsInternal; n++)
{
for (i = 0; i < channel_state[n].nFramesPerPacket; i++)
{
channel_state[n].VAD_flags[i] = psRangeDec.dec_bit_logp(1);
}
channel_state[n].LBRR_flag = psRangeDec.dec_bit_logp(1);
}
/* Decode LBRR flags */
for (n = 0; n < decControl.nChannelsInternal; n++)
{
Arrays.MemSetInt(channel_state[n].LBRR_flags, 0, SilkConstants.MAX_FRAMES_PER_PACKET);
if (channel_state[n].LBRR_flag != 0)
{
if (channel_state[n].nFramesPerPacket == 1)
{
channel_state[n].LBRR_flags[0] = 1;
}
else
{
LBRR_symbol = psRangeDec.dec_icdf(Tables.silk_LBRR_flags_iCDF_ptr[channel_state[n].nFramesPerPacket - 2], 8) + 1;
for (i = 0; i < channel_state[n].nFramesPerPacket; i++)
{
channel_state[n].LBRR_flags[i] = Inlines.silk_RSHIFT(LBRR_symbol, i) & 1;
}
}
}
}
if (lostFlag == DecoderAPIFlag.FLAG_DECODE_NORMAL)
{
/* Regular decoding: skip all LBRR data */
for (i = 0; i < channel_state[0].nFramesPerPacket; i++)
{
for (n = 0; n < decControl.nChannelsInternal; n++)
{
if (channel_state[n].LBRR_flags[i] != 0)
{
short[] pulses = new short[SilkConstants.MAX_FRAME_LENGTH];
int condCoding;
if (decControl.nChannelsInternal == 2 && n == 0)
{
Stereo.silk_stereo_decode_pred(psRangeDec, MS_pred_Q13);
if (channel_state[1].LBRR_flags[i] == 0)
{
BoxedValueInt decodeOnlyMiddleBoxed = new BoxedValueInt(decode_only_middle);
Stereo.silk_stereo_decode_mid_only(psRangeDec, decodeOnlyMiddleBoxed);
decode_only_middle = decodeOnlyMiddleBoxed.Val;
}
}
/* Use conditional coding if previous frame available */
if (i > 0 && (channel_state[n].LBRR_flags[i - 1] != 0))
{
condCoding = SilkConstants.CODE_CONDITIONALLY;
}
else
{
condCoding = SilkConstants.CODE_INDEPENDENTLY;
}
DecodeIndices.silk_decode_indices(channel_state[n], psRangeDec, i, 1, condCoding);
DecodePulses.silk_decode_pulses(psRangeDec, pulses, channel_state[n].indices.signalType,
channel_state[n].indices.quantOffsetType, channel_state[n].frame_length);
}
}
}
}
}
/* Get MS predictor index */
if (decControl.nChannelsInternal == 2)
{
if (lostFlag == DecoderAPIFlag.FLAG_DECODE_NORMAL ||
(lostFlag == DecoderAPIFlag.FLAG_DECODE_LBRR && channel_state[0].LBRR_flags[channel_state[0].nFramesDecoded] == 1))
{
Stereo.silk_stereo_decode_pred(psRangeDec, MS_pred_Q13);
/* For LBRR data, decode mid-only flag only if side-channel's LBRR flag is false */
if ((lostFlag == DecoderAPIFlag.FLAG_DECODE_NORMAL && channel_state[1].VAD_flags[channel_state[0].nFramesDecoded] == 0) ||
(lostFlag == DecoderAPIFlag.FLAG_DECODE_LBRR && channel_state[1].LBRR_flags[channel_state[0].nFramesDecoded] == 0))
{
BoxedValueInt decodeOnlyMiddleBoxed = new BoxedValueInt(decode_only_middle);
Stereo.silk_stereo_decode_mid_only(psRangeDec, decodeOnlyMiddleBoxed);
decode_only_middle = decodeOnlyMiddleBoxed.Val;
}
else
{
decode_only_middle = 0;
}
}
else
{
for (n = 0; n < 2; n++)
{
MS_pred_Q13[n] = psDec.sStereo.pred_prev_Q13[n];
}
}
}
/* Reset side channel decoder prediction memory for first frame with side coding */
if (decControl.nChannelsInternal == 2 && decode_only_middle == 0 && psDec.prev_decode_only_middle == 1)
{
Arrays.MemSetShort(psDec.channel_state[1].outBuf, 0, SilkConstants.MAX_FRAME_LENGTH + 2 * SilkConstants.MAX_SUB_FRAME_LENGTH);
Arrays.MemSetInt(psDec.channel_state[1].sLPC_Q14_buf, 0, SilkConstants.MAX_LPC_ORDER);
psDec.channel_state[1].lagPrev = 100;
psDec.channel_state[1].LastGainIndex = 10;
psDec.channel_state[1].prevSignalType = SilkConstants.TYPE_NO_VOICE_ACTIVITY;
psDec.channel_state[1].first_frame_after_reset = 1;
}
/* Check if the temp buffer fits into the output PCM buffer. If it fits,
* we can delay allocating the temp buffer until after the SILK peak stack
* usage. We need to use a < and not a <= because of the two extra samples. */
delay_stack_alloc = (decControl.internalSampleRate * decControl.nChannelsInternal
< decControl.API_sampleRate * decControl.nChannelsAPI) ? 1 : 0;
if (delay_stack_alloc != 0)
{
samplesOut_tmp = samplesOut;
samplesOut_tmp_ptrs[0] = samplesOut_ptr;
samplesOut_tmp_ptrs[1] = samplesOut_ptr + channel_state[0].frame_length + 2;
}
else
{
samplesOut1_tmp_storage1 = new short[decControl.nChannelsInternal * (channel_state[0].frame_length + 2)];
samplesOut_tmp = samplesOut1_tmp_storage1;
samplesOut_tmp_ptrs[0] = 0;
samplesOut_tmp_ptrs[1] = channel_state[0].frame_length + 2;
}
if (lostFlag == DecoderAPIFlag.FLAG_DECODE_NORMAL)
{
has_side = (decode_only_middle == 0) ? 1 : 0;
}
else
{
has_side = (psDec.prev_decode_only_middle == 0 ||
(decControl.nChannelsInternal == 2 &&
lostFlag == DecoderAPIFlag.FLAG_DECODE_LBRR &&
channel_state[1].LBRR_flags[channel_state[1].nFramesDecoded] == 1)) ? 1 : 0;
}
/* Call decoder for one frame */
for (n = 0; n < decControl.nChannelsInternal; n++)
{
if (n == 0 || (has_side != 0))
{
int FrameIndex;
int condCoding;
FrameIndex = channel_state[0].nFramesDecoded - n;
/* Use independent coding if no previous frame available */
if (FrameIndex <= 0)
{
condCoding = SilkConstants.CODE_INDEPENDENTLY;
}
else if (lostFlag == DecoderAPIFlag.FLAG_DECODE_LBRR)
{
condCoding = (channel_state[n].LBRR_flags[FrameIndex - 1] != 0) ? SilkConstants.CODE_CONDITIONALLY : SilkConstants.CODE_INDEPENDENTLY;
}
else if (n > 0 && (psDec.prev_decode_only_middle != 0))
{
/* If we skipped a side frame in this packet, we don't
* need LTP scaling; the LTP state is well-defined. */
condCoding = SilkConstants.CODE_INDEPENDENTLY_NO_LTP_SCALING;
}
else
{
condCoding = SilkConstants.CODE_CONDITIONALLY;
}
ret += channel_state[n].silk_decode_frame(psRangeDec, samplesOut_tmp, samplesOut_tmp_ptrs[n] + 2, nSamplesOutDec, lostFlag, condCoding);
}
else
{
Arrays.MemSetWithOffset <short>(samplesOut_tmp, 0, samplesOut_tmp_ptrs[n] + 2, nSamplesOutDec.Val);
}
channel_state[n].nFramesDecoded++;
}
if (decControl.nChannelsAPI == 2 && decControl.nChannelsInternal == 2)
{
/* Convert Mid/Side to Left/Right */
Stereo.silk_stereo_MS_to_LR(psDec.sStereo, samplesOut_tmp, samplesOut_tmp_ptrs[0], samplesOut_tmp, samplesOut_tmp_ptrs[1], MS_pred_Q13, channel_state[0].fs_kHz, nSamplesOutDec.Val);
}
else
{
/* Buffering */
Array.Copy(psDec.sStereo.sMid, 0, samplesOut_tmp, samplesOut_tmp_ptrs[0], 2);
Array.Copy(samplesOut_tmp, samplesOut_tmp_ptrs[0] + nSamplesOutDec.Val, psDec.sStereo.sMid, 0, 2);
}
/* Number of output samples */
nSamplesOut = Inlines.silk_DIV32(nSamplesOutDec.Val * decControl.API_sampleRate, Inlines.silk_SMULBB(channel_state[0].fs_kHz, 1000));
/* Set up pointers to temp buffers */
if (decControl.nChannelsAPI == 2)
{
samplesOut2_tmp = new short[nSamplesOut];
resample_out = samplesOut2_tmp;
resample_out_ptr = 0;
}
else
{
resample_out = samplesOut;
resample_out_ptr = samplesOut_ptr;
}
if (delay_stack_alloc != 0)
{
samplesOut1_tmp_storage2 = new short[decControl.nChannelsInternal * (channel_state[0].frame_length + 2)];
Array.Copy(samplesOut, samplesOut_ptr, samplesOut1_tmp_storage2, 0, decControl.nChannelsInternal * (channel_state[0].frame_length + 2));
samplesOut_tmp = samplesOut1_tmp_storage2;
samplesOut_tmp_ptrs[0] = 0;
samplesOut_tmp_ptrs[1] = channel_state[0].frame_length + 2;
}
for (n = 0; n < Inlines.silk_min(decControl.nChannelsAPI, decControl.nChannelsInternal); n++)
{
/* Resample decoded signal to API_sampleRate */
ret += Resampler.silk_resampler(channel_state[n].resampler_state, resample_out, resample_out_ptr, samplesOut_tmp, samplesOut_tmp_ptrs[n] + 1, nSamplesOutDec.Val);
/* Interleave if stereo output and stereo stream */
if (decControl.nChannelsAPI == 2)
{
int nptr = samplesOut_ptr + n;
for (i = 0; i < nSamplesOut; i++)
{
samplesOut[nptr + 2 * i] = resample_out[resample_out_ptr + i];
}
}
}
/* Create two channel output from mono stream */
if (decControl.nChannelsAPI == 2 && decControl.nChannelsInternal == 1)
{
if (stereo_to_mono != 0)
{
/* Resample right channel for newly collapsed stereo just in case
* we weren't doing collapsing when switching to mono */
ret += Resampler.silk_resampler(channel_state[1].resampler_state, resample_out, resample_out_ptr, samplesOut_tmp, samplesOut_tmp_ptrs[0] + 1, nSamplesOutDec.Val);
for (i = 0; i < nSamplesOut; i++)
{
samplesOut[samplesOut_ptr + 1 + 2 * i] = resample_out[resample_out_ptr + i];
}
}
else
{
for (i = 0; i < nSamplesOut; i++)
{
samplesOut[samplesOut_ptr + 1 + 2 * i] = samplesOut[samplesOut_ptr + 2 * i];
}
}
}
/* Export pitch lag, measured at 48 kHz sampling rate */
if (channel_state[0].prevSignalType == SilkConstants.TYPE_VOICED)
{
int[] mult_tab = { 6, 4, 3 };
decControl.prevPitchLag = channel_state[0].lagPrev * mult_tab[(channel_state[0].fs_kHz - 8) >> 2];
}
else
{
decControl.prevPitchLag = 0;
}
if (lostFlag == DecoderAPIFlag.FLAG_PACKET_LOST)
{
/* On packet loss, remove the gain clamping to prevent having the energy "bounce back"
* if we lose packets when the energy is going down */
for (i = 0; i < psDec.nChannelsInternal; i++)
{
psDec.channel_state[i].LastGainIndex = 10;
}
}
else
{
psDec.prev_decode_only_middle = decode_only_middle;
}
return(ret);
}
