From f3fdae3f8f4ae54b17f925f81a5e9d4b3589269b Mon Sep 17 00:00:00 2001 From: Carl Hetherington Date: Sat, 29 Sep 2012 23:41:25 +0100 Subject: Entirely untested resampling to fix 24fps drop-frame. --- src/lib/decoder.cc | 105 +---------------------------------------------------- 1 file changed, 2 insertions(+), 103 deletions(-) (limited to 'src/lib/decoder.cc') diff --git a/src/lib/decoder.cc b/src/lib/decoder.cc index 973582ca4..b7aca764d 100644 --- a/src/lib/decoder.cc +++ b/src/lib/decoder.cc @@ -69,9 +69,6 @@ Decoder::Decoder (boost::shared_ptr s, boost::shared_ptraudio_sample_rate != dcp_audio_sample_rate (_fs->audio_sample_rate)) { -#if HAVE_SWRESAMPLE - _swr_context = swr_alloc_set_opts ( - 0, - audio_channel_layout(), - audio_sample_format(), - dcp_audio_sample_rate (_fs->audio_sample_rate), - audio_channel_layout(), - audio_sample_format(), - _fs->audio_sample_rate, - 0, 0 - ); - - swr_init (_swr_context); -#else - throw DecodeError ("Cannot resample audio as libswresample is not present"); -#endif - } else { -#if HAVE_SWRESAMPLE - _swr_context = 0; -#endif - } - _delay_in_bytes = _fs->audio_delay * _fs->audio_sample_rate * _fs->audio_channels * _fs->bytes_per_sample() / 1000; delete _delay_line; _delay_line = new DelayLine (_delay_in_bytes); @@ -125,35 +99,6 @@ Decoder::process_begin () void Decoder::process_end () { -#if HAVE_SWRESAMPLE - if (_swr_context) { - - int mop = 0; - while (1) { - uint8_t buffer[256 * _fs->bytes_per_sample() * _fs->audio_channels]; - uint8_t* out[1] = { - buffer - }; - - int const frames = swr_convert (_swr_context, out, 256, 0, 0); - - if (frames < 0) { - throw DecodeError ("could not run sample-rate converter"); - } - - if (frames == 0) { - break; - } - - mop += frames; - int available = _delay_line->feed (buffer, frames * _fs->audio_channels * _fs->bytes_per_sample()); - Audio (buffer, available); - } - - swr_free (&_swr_context); - } -#endif - if (_delay_in_bytes < 0) { uint8_t remainder[-_delay_in_bytes]; _delay_line->get_remaining (remainder); @@ -166,7 +111,7 @@ Decoder::process_end () */ int64_t const audio_short_by_frames = - ((int64_t) decoding_frames() * dcp_audio_sample_rate (_fs->audio_sample_rate) / _fs->frames_per_second) + ((int64_t) decoding_frames() * _fs->audio_sample_rate / _fs->frames_per_second) - _audio_frames_processed; if (audio_short_by_frames >= 0) { @@ -240,16 +185,9 @@ Decoder::pass () void Decoder::process_audio (uint8_t* data, int size) { - /* Here's samples per channel */ + /* Samples per channel */ int const samples = size / _fs->bytes_per_sample(); -#if HAVE_SWRESAMPLE - /* And here's frames (where 1 frame is a collection of samples, 1 for each channel, - so for 5.1 a frame would be 6 samples) - */ - int const frames = samples / _fs->audio_channels; -#endif - /* Maybe apply gain */ if (_fs->audio_gain != 0) { float const linear_gain = pow (10, _fs->audio_gain / 20); @@ -282,51 +220,12 @@ Decoder::process_audio (uint8_t* data, int size) } } - /* This is a buffer we might use if we are sample-rate converting; - it will need freeing if so. - */ - uint8_t* out_buffer = 0; - - /* Maybe sample-rate convert */ -#if HAVE_SWRESAMPLE - if (_swr_context) { - - uint8_t const * in[2] = { - data, - 0 - }; - - /* Compute the resampled frame count and add 32 for luck */ - int const out_buffer_size_frames = ceil (frames * float (dcp_audio_sample_rate (_fs->audio_sample_rate)) / _fs->audio_sample_rate) + 32; - int const out_buffer_size_bytes = out_buffer_size_frames * _fs->audio_channels * _fs->bytes_per_sample(); - out_buffer = new uint8_t[out_buffer_size_bytes]; - - uint8_t* out[2] = { - out_buffer, - 0 - }; - - /* Resample audio */ - int out_frames = swr_convert (_swr_context, out, out_buffer_size_frames, in, frames); - if (out_frames < 0) { - throw DecodeError ("could not run sample-rate converter"); - } - - /* And point our variables at the resampled audio */ - data = out_buffer; - size = out_frames * _fs->audio_channels * _fs->bytes_per_sample(); - } -#endif - /* Update the number of audio frames we've pushed to the encoder */ _audio_frames_processed += size / (_fs->audio_channels * _fs->bytes_per_sample ()); /* Push into the delay line and then tell the world what we've got */ int available = _delay_line->feed (data, size); Audio (data, available); - - /* Delete the sample-rate conversion buffer, if it exists */ - delete[] out_buffer; } /** Called by subclasses to tell the world that some video data is ready. -- cgit v1.2.3 From 0f154f43bd0c88d1615e455bd8a169826a08c086 Mon Sep 17 00:00:00 2001 From: Carl Hetherington Date: Mon, 1 Oct 2012 19:51:36 +0100 Subject: Various fixes to resampling. --- src/lib/decoder.cc | 13 +++++++---- src/lib/film_state.cc | 21 ++++++++++++++++++ src/lib/film_state.h | 1 + src/lib/j2k_wav_encoder.cc | 54 +++++++++++++++------------------------------- src/lib/j2k_wav_encoder.h | 2 -- 5 files changed, 48 insertions(+), 43 deletions(-) (limited to 'src/lib/decoder.cc') diff --git a/src/lib/decoder.cc b/src/lib/decoder.cc index b7aca764d..213ff9dd4 100644 --- a/src/lib/decoder.cc +++ b/src/lib/decoder.cc @@ -111,18 +111,23 @@ Decoder::process_end () */ int64_t const audio_short_by_frames = - ((int64_t) decoding_frames() * _fs->audio_sample_rate / _fs->frames_per_second) + ((int64_t) decoding_frames() * _fs->target_sample_rate() / _fs->frames_per_second) - _audio_frames_processed; if (audio_short_by_frames >= 0) { - int bytes = audio_short_by_frames * _fs->audio_channels * _fs->bytes_per_sample(); + + stringstream s; + s << "Adding " << audio_short_by_frames << " frames of silence to the end."; + _log->log (s.str ()); + + int64_t bytes = audio_short_by_frames * _fs->audio_channels * _fs->bytes_per_sample(); - int const silence_size = 64 * 1024; + int64_t const silence_size = 64 * 1024; uint8_t silence[silence_size]; memset (silence, 0, silence_size); while (bytes) { - int const t = min (bytes, silence_size); + int64_t const t = min (bytes, silence_size); Audio (silence, t); bytes -= t; } diff --git a/src/lib/film_state.cc b/src/lib/film_state.cc index e472434ce..0c1ac87dc 100644 --- a/src/lib/film_state.cc +++ b/src/lib/film_state.cc @@ -35,6 +35,7 @@ #include "format.h" #include "dcp_content_type.h" #include "util.h" +#include "exceptions.h" using namespace std; using namespace boost; @@ -278,3 +279,23 @@ FilmState::bytes_per_sample () const return 0; } + +int +FilmState::target_sample_rate () const +{ + double t = dcp_audio_sample_rate (audio_sample_rate); + if (rint (frames_per_second) != frames_per_second) { + if (fabs (frames_per_second - 23.976) < 1e-6) { + /* 24fps drop-frame ie 24 * 1000 / 1001 frames per second; + hence we need to resample the audio to dcp_audio_sample_rate * 1000 / 1001 + so that when we play it back at dcp_audio_sample_rate it is sped up + by the same amount that the video is + */ + t *= double(1000) / 1001; + } else { + throw EncodeError ("unknown fractional frame rate"); + } + } + + return rint (t); +} diff --git a/src/lib/film_state.h b/src/lib/film_state.h index 12d44cdce..8dc0ce11b 100644 --- a/src/lib/film_state.h +++ b/src/lib/film_state.h @@ -80,6 +80,7 @@ public: int thumb_frame (int) const; int bytes_per_sample () const; + int target_sample_rate () const; void write_metadata (std::ofstream &) const; void read_metadata (std::string, std::string); diff --git a/src/lib/j2k_wav_encoder.cc b/src/lib/j2k_wav_encoder.cc index 241639400..9b25717ef 100644 --- a/src/lib/j2k_wav_encoder.cc +++ b/src/lib/j2k_wav_encoder.cc @@ -219,14 +219,14 @@ J2KWAVEncoder::process_begin (int64_t audio_channel_layout, AVSampleFormat audio #ifdef HAVE_SWRESAMPLE stringstream s; - s << "Will resample audio from " << _fs->audio_sample_rate << " to " << target_sample_rate(); + s << "Will resample audio from " << _fs->audio_sample_rate << " to " << _fs->target_sample_rate(); _log->log (s.str ()); _swr_context = swr_alloc_set_opts ( 0, audio_channel_layout, audio_sample_format, - target_sample_rate(), + _fs->target_sample_rate(), audio_channel_layout, audio_sample_format, _fs->audio_sample_rate, @@ -303,11 +303,11 @@ J2KWAVEncoder::process_end () #if HAVE_SWRESAMPLE if (_swr_context) { - int mop = 0; while (1) { uint8_t buffer[256 * _fs->bytes_per_sample() * _fs->audio_channels]; - uint8_t* out[1] = { - buffer + uint8_t* out[2] = { + buffer, + 0 }; int const frames = swr_convert (_swr_context, out, 256, 0, 0); @@ -320,8 +320,7 @@ J2KWAVEncoder::process_end () break; } - mop += frames; - write_audio (buffer, frames); + write_audio (buffer, frames * _fs->bytes_per_sample() * _fs->audio_channels); } swr_free (&_swr_context); @@ -365,7 +364,7 @@ J2KWAVEncoder::process_audio (uint8_t* data, int size) int const frames = samples / _fs->audio_channels; /* Compute the resampled frame count and add 32 for luck */ - int const out_buffer_size_frames = ceil (frames * target_sample_rate() / _fs->audio_sample_rate) + 32; + int const out_buffer_size_frames = ceil (frames * _fs->target_sample_rate() / _fs->audio_sample_rate) + 32; int const out_buffer_size_bytes = out_buffer_size_frames * _fs->audio_channels * _fs->bytes_per_sample(); out_buffer = new uint8_t[out_buffer_size_bytes]; @@ -375,7 +374,7 @@ J2KWAVEncoder::process_audio (uint8_t* data, int size) }; /* Resample audio */ - int out_frames = swr_convert (_swr_context, out, out_buffer_size_frames, in, size); + int out_frames = swr_convert (_swr_context, out, out_buffer_size_frames, in, frames); if (out_frames < 0) { throw EncodeError ("could not run sample-rate converter"); } @@ -395,12 +394,12 @@ J2KWAVEncoder::process_audio (uint8_t* data, int size) void J2KWAVEncoder::write_audio (uint8_t* data, int size) { - /* Size of a sample in bytes */ - int const sample_size = 2; - - /* XXX: we are assuming that sample_size is right, the _deinterleave_buffer_size is a multiple - of the sample size and that data_size is a multiple of _fs->audio_channels * sample_size. + /* XXX: we are assuming that the _deinterleave_buffer_size is a multiple + of the sample size and that size is a multiple of _fs->audio_channels * sample_size. */ + + assert ((size % (_fs->audio_channels * _fs->bytes_per_sample())) == 0); + assert ((_deinterleave_buffer_size % _fs->bytes_per_sample()) == 0); /* XXX: this code is very tricksy and it must be possible to make it simpler ... */ @@ -412,17 +411,17 @@ J2KWAVEncoder::write_audio (uint8_t* data, int size) /* How many bytes of the deinterleaved data to do this time */ int this_time = min (remaining / _fs->audio_channels, _deinterleave_buffer_size); for (int i = 0; i < _fs->audio_channels; ++i) { - for (int j = 0; j < this_time; j += sample_size) { - for (int k = 0; k < sample_size; ++k) { + for (int j = 0; j < this_time; j += _fs->bytes_per_sample()) { + for (int k = 0; k < _fs->bytes_per_sample(); ++k) { int const to = j + k; - int const from = position + (i * sample_size) + (j * _fs->audio_channels) + k; + int const from = position + (i * _fs->bytes_per_sample()) + (j * _fs->audio_channels) + k; _deinterleave_buffer[to] = data[from]; } } switch (_fs->audio_sample_format) { case AV_SAMPLE_FMT_S16: - sf_write_short (_sound_files[i], (const short *) _deinterleave_buffer, this_time / sample_size); + sf_write_short (_sound_files[i], (const short *) _deinterleave_buffer, this_time / _fs->bytes_per_sample()); break; default: throw EncodeError ("unknown audio sample format"); @@ -434,22 +433,3 @@ J2KWAVEncoder::write_audio (uint8_t* data, int size) } } -int -J2KWAVEncoder::target_sample_rate () const -{ - double t = dcp_audio_sample_rate (_fs->audio_sample_rate); - if (rint (_fs->frames_per_second) != _fs->frames_per_second) { - if (_fs->frames_per_second == 23.976) { - /* 24fps drop-frame ie 24 * 1000 / 1001 frames per second; - hence we need to resample the audio to dcp_audio_sample_rate * 1000 / 1001 - so that when we play it back at dcp_audio_sample_rate it is sped up - by the same amount that the video is - */ - t *= double(1000) / 1001; - } else { - throw EncodeError ("unknown fractional frame rate"); - } - } - - return rint (t); -} diff --git a/src/lib/j2k_wav_encoder.h b/src/lib/j2k_wav_encoder.h index 3f01ac480..e11358c2c 100644 --- a/src/lib/j2k_wav_encoder.h +++ b/src/lib/j2k_wav_encoder.h @@ -55,8 +55,6 @@ public: private: - int target_sample_rate () const; - void write_audio (uint8_t* data, int size); void encoder_thread (ServerDescription *); void close_sound_files (); -- cgit v1.2.3