public ALACWriter(string path, Stream IO, AudioPCMConfig pcm)
{
_pcm = pcm;
if (_pcm.BitsPerSample != 16)
throw new Exception("Bits per sample must be 16.");
if (_pcm.ChannelCount != 2)
throw new Exception("ChannelCount must be 2.");
_path = path;
_IO = IO;
_pathGiven = _IO == null;
if (_IO != null && !_IO.CanSeek)
throw new NotSupportedException("stream doesn't support seeking");
samplesBuffer = new int[Alac.MAX_BLOCKSIZE * (_pcm.ChannelCount == 2 ? 5 : _pcm.ChannelCount)];
residualBuffer = new int[Alac.MAX_BLOCKSIZE * (_pcm.ChannelCount == 2 ? 6 : _pcm.ChannelCount + 1)];
windowBuffer = new float[Alac.MAX_BLOCKSIZE * 2 * Alac.MAX_LPC_WINDOWS];
eparams.set_defaults(_compressionLevel);
eparams.padding_size = 4096;
crc8 = new Crc8();
crc16 = new Crc16();
frame = new ALACFrame(_pcm.ChannelCount == 2 ? 5 : _pcm.ChannelCount);
chunk_pos = new List<int>();
}
unsafe void encode_residual_pass2(ALACFrame frame, int ch)
{
encode_residual(frame, ch, 2, estimate_best_window(frame, ch));
}
unsafe int estimate_best_window(ALACFrame frame, int ch)
{
if (_windowcount == 1)
return 0;
switch (eparams.window_method)
{
case WindowMethod.Estimate:
{
int best_window = -1;
double best_error = 0;
int order = 2;
for (int i = 0; i < _windowcount; i++)
{
frame.subframes[ch].lpc_ctx[i].GetReflection(order, frame.subframes[ch].samples, frame.blocksize, frame.window_buffer + i * Alac.MAX_BLOCKSIZE * 2);
double err = frame.subframes[ch].lpc_ctx[i].prediction_error[order - 1] / frame.subframes[ch].lpc_ctx[i].autocorr_values[0];
if (best_window == -1 || best_error > err)
{
best_window = i;
best_error = err;
}
}
return best_window;
}
case WindowMethod.Evaluate:
encode_residual_pass1(frame, ch, -1);
return frame.subframes[ch].best.window;
case WindowMethod.Search:
return -1;
}
return -1;
}
unsafe void output_frame_header(ALACFrame frame, BitWriter bitwriter)
{
bitwriter.writebits(3, _pcm.ChannelCount - 1);
bitwriter.writebits(16, 0);
bitwriter.writebits(1, frame.blocksize != eparams.block_size ? 1 : 0); // sample count is in the header
bitwriter.writebits(2, 0); // wasted bytes
bitwriter.writebits(1, frame.type == FrameType.Verbatim ? 1 : 0); // is verbatim
if (frame.blocksize != eparams.block_size)
bitwriter.writebits(32, frame.blocksize);
if (frame.type != FrameType.Verbatim)
{
bitwriter.writebits(8, frame.interlacing_shift);
bitwriter.writebits(8, frame.interlacing_leftweight);
for (int ch = 0; ch < _pcm.ChannelCount; ch++)
{
bitwriter.writebits(4, 0); // prediction type
bitwriter.writebits(4, frame.subframes[ch].best.shift);
bitwriter.writebits(3, frame.subframes[ch].best.ricemodifier);
bitwriter.writebits(5, frame.subframes[ch].best.order);
if (frame.subframes[ch].best.order != 31)
for (int c = 0; c < frame.subframes[ch].best.order; c++)
bitwriter.writebits_signed(16, frame.subframes[ch].best.coefs[c]);
}
}
}
unsafe void encode_residual_pass1(ALACFrame frame, int ch, int best_window)
{
int max_prediction_order = eparams.max_prediction_order;
int estimation_depth = eparams.estimation_depth;
int min_modifier = eparams.min_modifier;
int adaptive_passes = eparams.adaptive_passes;
eparams.max_prediction_order = Math.Min(8,eparams.max_prediction_order);
eparams.estimation_depth = 1;
eparams.min_modifier = eparams.max_modifier;
eparams.adaptive_passes = 0;
encode_residual(frame, ch, 1, best_window);
eparams.max_prediction_order = max_prediction_order;
eparams.estimation_depth = estimation_depth;
eparams.min_modifier = min_modifier;
eparams.adaptive_passes = adaptive_passes;
}
unsafe void encode_residual(ALACFrame frame, int ch, int pass, int best_window)
{
int* smp = frame.subframes[ch].samples;
int i, n = frame.blocksize;
int bps = _pcm.BitsPerSample + _pcm.ChannelCount - 1;
// FIXED
//if (0 == (2 & frame.subframes[ch].done_fixed) && (pass != 1 || n < eparams.max_prediction_order))
//{
// frame.subframes[ch].done_fixed |= 2;
// frame.current.order = 31;
// frame.current.window = -1;
// alac_encode_residual_31(frame.current.residual, frame.subframes[ch].samples, frame.blocksize);
// frame.current.size = (uint)(alac_entropy_coder(frame.current.residual, frame.blocksize, bps, out frame.current.ricemodifier) + 16);
// frame.ChooseBestSubframe(ch);
//}
//if (0 == (1 & frame.subframes[ch].done_fixed) && (pass != 1 || n < eparams.max_prediction_order))
//{
// frame.subframes[ch].done_fixed |= 1;
// frame.current.order = 0;
// frame.current.window = -1;
// alac_encode_residual_0(frame.current.residual, frame.subframes[ch].samples, frame.blocksize);
// frame.current.size = (uint)(alac_entropy_coder(frame.current.residual, frame.blocksize, bps, out frame.current.ricemodifier) + 16);
// frame.ChooseBestSubframe(ch);
//}
// LPC
if (n < eparams.max_prediction_order)
return;
float* lpcs = stackalloc float[lpc.MAX_LPC_ORDER * lpc.MAX_LPC_ORDER];
int min_order = eparams.min_prediction_order;
int max_order = eparams.max_prediction_order;
for (int iWindow = 0; iWindow < _windowcount; iWindow++)
{
if (best_window != -1 && iWindow != best_window)
continue;
LpcContext lpc_ctx = frame.subframes[ch].lpc_ctx[iWindow];
lpc_ctx.GetReflection(max_order, smp, n, frame.window_buffer + iWindow * Alac.MAX_BLOCKSIZE * 2);
lpc_ctx.ComputeLPC(lpcs);
lpc_ctx.SortOrdersAkaike(frame.blocksize, eparams.estimation_depth, max_order, 5.0, 1.0/18);
for (i = 0; i < eparams.estimation_depth && i < max_order; i++)
encode_residual_lpc_sub(frame, lpcs, iWindow, lpc_ctx.best_orders[i], ch);
}
}
unsafe void encode_residual_lpc_sub(ALACFrame frame, float* lpcs, int iWindow, int order, int ch)
{
// check if we already calculated with this order, window and precision
if ((frame.subframes[ch].lpc_ctx[iWindow].done_lpcs[eparams.adaptive_passes] & (1U << (order - 1))) == 0)
{
frame.subframes[ch].lpc_ctx[iWindow].done_lpcs[eparams.adaptive_passes] |= (1U << (order - 1));
uint cbits = 15U;
frame.current.order = order;
frame.current.window = iWindow;
int bps = _pcm.BitsPerSample + _pcm.ChannelCount - 1;
int* coefs = stackalloc int[lpc.MAX_LPC_ORDER];
//if (frame.subframes[ch].best.order == order && frame.subframes[ch].best.window == iWindow)
//{
// frame.current.shift = frame.subframes[ch].best.shift;
// for (int i = 0; i < frame.current.order; i++)
// frame.current.coefs[i] = frame.subframes[ch].best.coefs_adapted[i];
//}
//else
{
lpc.quantize_lpc_coefs(lpcs + (frame.current.order - 1) * lpc.MAX_LPC_ORDER,
frame.current.order, cbits, coefs, out frame.current.shift, 15, 1);
if (frame.current.shift < 0 || frame.current.shift > 15)
throw new Exception("negative shift");
for (int i = 0; i < frame.current.order; i++)
frame.current.coefs[i] = coefs[i];
}
for (int i = 0; i < frame.current.order; i++)
coefs[i] = frame.current.coefs[frame.current.order - 1 - i];
for (int i = frame.current.order; i < lpc.MAX_LPC_ORDER; i++)
coefs[i] = 0;
alac_encode_residual(frame.current.residual, frame.subframes[ch].samples, frame.blocksize,
frame.current.order, coefs, frame.current.shift, bps);
for (int i = 0; i < frame.current.order; i++)
frame.current.coefs_adapted[i] = coefs[frame.current.order - 1 - i];
for (int adaptive_pass = 0; adaptive_pass < eparams.adaptive_passes; adaptive_pass++)
{
for (int i = 0; i < frame.current.order; i++)
frame.current.coefs[i] = frame.current.coefs_adapted[i];
alac_encode_residual(frame.current.residual, frame.subframes[ch].samples, frame.blocksize,
frame.current.order, coefs, frame.current.shift, bps);
for (int i = 0; i < frame.current.order; i++)
frame.current.coefs_adapted[i] = coefs[frame.current.order - 1 - i];
}
frame.current.size = (uint)(alac_entropy_estimate(frame.current.residual, frame.blocksize, bps, eparams.max_modifier) + 16 + 16 * order);
frame.ChooseBestSubframe(ch);
}
}
unsafe void encode_estimated_frame(ALACFrame frame)
{
switch (eparams.stereo_method)
{
case StereoMethod.Estimate:
for (int ch = 0; ch < _pcm.ChannelCount; ch++)
{
frame.subframes[ch].best.size = AudioSamples.UINT32_MAX;
encode_residual_pass2(frame, ch);
}
break;
case StereoMethod.Evaluate:
for (int ch = 0; ch < _pcm.ChannelCount; ch++)
encode_residual_pass2(frame, ch);
break;
case StereoMethod.Search:
break;
}
}
unsafe uint measure_frame_size(ALACFrame frame, bool do_midside)
{
// crude estimation of header/footer size
uint total = 16 + 3;
if (do_midside)
{
uint bitsBest = frame.subframes[0].best.size + frame.subframes[1].best.size;
frame.interlacing_leftweight = 0;
frame.interlacing_shift = 0;
if (bitsBest > frame.subframes[3].best.size + frame.subframes[0].best.size) // leftside
{
bitsBest = frame.subframes[3].best.size + frame.subframes[0].best.size;
frame.interlacing_leftweight = 1;
frame.interlacing_shift = 0;
}
if (bitsBest > frame.subframes[3].best.size + frame.subframes[2].best.size) // midside
{
bitsBest = frame.subframes[3].best.size + frame.subframes[2].best.size;
frame.interlacing_leftweight = 1;
frame.interlacing_shift = 1;
}
if (bitsBest > frame.subframes[3].best.size + frame.subframes[4].best.size) // rightside
{
bitsBest = frame.subframes[3].best.size + frame.subframes[4].best.size;
frame.interlacing_leftweight = 1;
frame.interlacing_shift = 31;
}
return total + bitsBest;
}
for (int ch = 0; ch < _pcm.ChannelCount; ch++)
total += frame.subframes[ch].best.size;
return total;
}
unsafe void estimate_frame(ALACFrame frame, bool do_midside)
{
int subframes = do_midside ? 5 : _pcm.ChannelCount;
switch (eparams.stereo_method)
{
case StereoMethod.Estimate:
for (int ch = 0; ch < subframes; ch++)
{
LpcContext lpc_ctx = frame.subframes[ch].lpc_ctx[0];
int stereo_order = Math.Min(8, eparams.max_prediction_order);
double alpha = 1.5; // 4.5 + eparams.max_prediction_order / 10.0;
lpc_ctx.GetReflection(stereo_order, frame.subframes[ch].samples, frame.blocksize, frame.window_buffer);
lpc_ctx.SortOrdersAkaike(frame.blocksize, 1, stereo_order, alpha, 0);
frame.subframes[ch].best.size = (uint)Math.Max(0, lpc_ctx.Akaike(frame.blocksize, lpc_ctx.best_orders[0], alpha, 0));
}
break;
case StereoMethod.Evaluate:
for (int ch = 0; ch < subframes; ch++)
encode_residual_pass1(frame, ch, 0);
break;
case StereoMethod.Search:
for (int ch = 0; ch < subframes; ch++)
encode_residual_pass2(frame, ch);
break;
}
}