diff options
author | Peter Belkner <pbelkner@snafu.de> | 2011-05-07 02:23:28 +0200 |
---|---|---|
committer | Michael Niedermayer <michaelni@gmx.at> | 2011-05-07 03:27:38 +0200 |
commit | d62bf5d4e73250295c0a652e151498c5b19cbd63 (patch) | |
tree | 08d484e42fea8a72e8db42b3c5f852e6f39ad1d6 /libavformat/mp3enc.c | |
parent | c34cb130b6bf3e957484fd236c1ba477720882c0 (diff) | |
download | ffmpeg-d62bf5d4e73250295c0a652e151498c5b19cbd63.tar.gz |
mp3enc: mux a XING header
The patch below provides exactly that to the MP3 muxer. A XING header
containing
* the numer of frames,
* the size, and
* a TOC
is generated.
It's based on an idea by Anton Khirnov (restricted to the number of
frames) found at
http://patches.ffmpeg.org/patch/1891/
The TOC is generated as found in lame's "VbrTag.c".
According to my tests the following reproduces the number of frames, the
size and the TOC in "c.mp3" from "b.mp3" (except a shift due to shorter
XING header generated by FFmpeg):
lame -V2 a.wav b.mp3
ffmpeg -i b.mp3 -acodec copy -y c.mp3
Diffstat (limited to 'libavformat/mp3enc.c')
-rw-r--r-- | libavformat/mp3enc.c | 229 |
1 files changed, 226 insertions, 3 deletions
diff --git a/libavformat/mp3enc.c b/libavformat/mp3enc.c index 9b02050178..8150b781ed 100644 --- a/libavformat/mp3enc.c +++ b/libavformat/mp3enc.c @@ -27,6 +27,9 @@ #include "libavutil/avstring.h" #include "libavutil/intreadwrite.h" #include "libavutil/opt.h" +#include "libavcodec/mpegaudiodata.h" +#include "libavcodec/mpegaudiodecheader.h" +#include "libavformat/avio_internal.h" static int id3v1_set_string(AVFormatContext *s, const char *key, uint8_t *buf, int buf_size) @@ -127,7 +130,7 @@ static int id3v2_put_ttag(AVFormatContext *s, const char *str1, const char *str2 return len + ID3v2_HEADER_SIZE; } -static int mp3_write_trailer(struct AVFormatContext *s) +static int mp2_write_trailer(struct AVFormatContext *s) { uint8_t buf[ID3v1_TAG_SIZE]; @@ -150,14 +153,29 @@ AVOutputFormat ff_mp2_muxer = { CODEC_ID_NONE, NULL, ff_raw_write_packet, - mp3_write_trailer, + mp2_write_trailer, }; #endif #if CONFIG_MP3_MUXER +#define VBR_NUM_BAGS 400 +#define VBR_TOC_SIZE 100 typedef struct MP3Context { const AVClass *class; int id3v2_version; + struct xing_header { + int64_t offset; + int32_t frames; + int32_t size; + /* following lame's "VbrTag.c". */ + struct xing_toc { + uint32_t want; + uint32_t seen; + uint32_t pos; + uint64_t sum; + uint64_t bag[VBR_NUM_BAGS]; + } toc; + } xing_header; } MP3Context; static const AVOption options[] = { @@ -188,6 +206,156 @@ static int id3v2_check_write_tag(AVFormatContext *s, AVMetadataTag *t, const cha return -1; } +static const int64_t xing_offtbl[2][2] = {{32, 17}, {17,9}}; +static const uint32_t XING = MKBETAG('X', 'i', 'n', 'g'); +#ifdef FILTER_VBR_HEADERS +static const uint32_t INFO = MKBETAG('I', 'n', 'f', 'o'); +static const uint32_t VBRI = MKBETAG('V', 'B', 'R', 'I'); +#endif + +/* + * Write an empty XING header and initialize respective data. + */ +static int mp3_write_xing(AVFormatContext *s) +{ + AVCodecContext *codec = s->streams[0]->codec; + MP3Context *mp3 = s->priv_data; + int bitrate_idx = 3; + int64_t xing_offset; + int32_t mask, header; + MPADecodeHeader c; + int srate_idx, i, channels; + int needed; + + for (i = 0; i < FF_ARRAY_ELEMS(ff_mpa_freq_tab); i++) + if (ff_mpa_freq_tab[i] == codec->sample_rate) { + srate_idx = i; + break; + } + if (i == FF_ARRAY_ELEMS(ff_mpa_freq_tab)) { + av_log(s, AV_LOG_ERROR, "Unsupported sample rate.\n"); + return -1; + } + + switch (codec->channels) { + case 1: channels = MPA_MONO; break; + case 2: channels = MPA_STEREO; break; + default: av_log(s, AV_LOG_ERROR, "Unsupported number of channels.\n"); return -1; + } + + /* dummy MPEG audio header */ + header = 0xff << 24; // sync + header |= (0x7 << 5 | 0x3 << 3 | 0x1 << 1 | 0x1) << 16; // sync/mpeg-1/layer 3/no crc*/ + header |= (srate_idx << 2) << 8; + header |= channels << 6; + + for (;;) { + if (15 == bitrate_idx) + return -1; + + mask = (bitrate_idx << 4) << 8; + header |= mask; + ff_mpegaudio_decode_header(&c, header); + xing_offset=xing_offtbl[c.lsf == 1][c.nb_channels == 1]; + needed = 4 // header + + xing_offset + + 4 // xing tag + + 4 // frames/size/toc flags + + 4 // frames + + 4 // size + + VBR_TOC_SIZE; // toc + + if (needed <= c.frame_size) + break; + + header &= ~mask; + ++bitrate_idx; + } + + avio_wb32(s->pb, header); + ffio_fill(s->pb, 0, xing_offset); + avio_wb32(s->pb, XING); + avio_wb32(s->pb, 0x01 | 0x02 | 0x04); // frames/size/toc + + mp3->xing_header.offset = avio_tell(s->pb); + mp3->xing_header.size = c.frame_size; + mp3->xing_header.toc.want=1; + mp3->xing_header.toc.seen=0; + mp3->xing_header.toc.pos=0; + mp3->xing_header.toc.sum=0; + + avio_wb32(s->pb, 0); // frames + avio_wb32(s->pb, 0); // size + + // toc + for (i = 0; i < VBR_TOC_SIZE; ++i) + avio_w8(s->pb, (uint8_t)(255 * i / VBR_TOC_SIZE)); + + ffio_fill(s->pb, 0, c.frame_size - needed); + avio_flush(s->pb); + + return 0; +} + +/* + * Add a frame to XING data. + * Following lame's "VbrTag.c". + */ +static void mp3_xing_add_frame(AVFormatContext *s, AVPacket *pkt, MPADecodeHeader *c) +{ + MP3Context *mp3 = s->priv_data; + struct xing_header *xing_header = &mp3->xing_header; + struct xing_toc *toc = &xing_header->toc; + int i; + + ++xing_header->frames; + xing_header->size += pkt->size; + toc->sum += c->bit_rate / 1000; + + if (toc->want == ++toc->seen) { + toc->bag[toc->pos] = toc->sum; + + if (VBR_NUM_BAGS == ++toc->pos) { + /* shrink table to half size by throwing away each second bag. */ + for (i = 1; i < VBR_NUM_BAGS; i += 2) + toc->bag[i >> 1] = toc->bag[i]; + + /* double wanted amount per bag. */ + toc->want <<= 1; + /* adjust current position to half of table size. */ + toc->pos >>= 1; + } + + toc->seen = 0; + } +} + +static void mp3_fix_xing(AVFormatContext *s) +{ + MP3Context *mp3 = s->priv_data; + struct xing_header *xing_header = &mp3->xing_header; + struct xing_toc *toc = &xing_header->toc; + double scale = (double)toc->pos / (double)VBR_TOC_SIZE; + int i; + + avio_flush(s->pb); + avio_seek(s->pb, xing_header->offset, SEEK_SET); + avio_wb32(s->pb, xing_header->frames); + avio_wb32(s->pb, xing_header->size); + + avio_w8(s->pb, 0); // first toc entry has to be zero. + + for (i = 1; i < VBR_TOC_SIZE; ++i) { + int j = (int)floor(scale * i); + int seek_point = (int)floor(256.0 * toc->bag[j] / toc->sum); + + avio_w8(s->pb, (uint8_t)(seek_point < 256 ? seek_point : 255)); + } + + avio_flush(s->pb); + avio_seek(s->pb, 0, SEEK_END); +} + /** * Write an ID3v2 header at beginning of stream */ @@ -236,6 +404,61 @@ static int mp3_write_header(struct AVFormatContext *s) id3v2_put_size(s, totlen); avio_seek(s->pb, cur_pos, SEEK_SET); + if (s->pb->seekable) + mp3_write_xing(s); + + return 0; +} + +static int mp3_write_packet(AVFormatContext *s, AVPacket *pkt) +{ + if (! pkt || ! pkt->data || pkt->size < 4) + return ff_raw_write_packet(s, pkt); + else { + MP3Context *mp3 = s->priv_data; + MPADecodeHeader c; +#ifdef FILTER_VBR_HEADERS + int base; +#endif + + ff_mpegaudio_decode_header(&c, AV_RB32(pkt->data)); + +#ifdef FILTER_VBR_HEADERS + /* filter out XING and INFO headers. */ + base = 4 + xing_offtbl[c.lsf == 1][c.nb_channels == 1]; + + if (base + 4 <= pkt->size) { + uint32_t v = AV_RB32(pkt->data + base); + + if (XING == v || INFO == v) + return 0; + } + + /* filter out VBRI headers. */ + base = 4 + 32; + + if (base + 4 <= pkt->size && VBRI == AV_RB32(pkt->data + base)) + return 0; +#endif + + if (0 < mp3->xing_header.offset) + mp3_xing_add_frame(s, pkt, &c); + + return ff_raw_write_packet(s, pkt); + } +} + +static int mp3_write_trailer(AVFormatContext *s) +{ + MP3Context *mp3 = s->priv_data; + int ret=mp2_write_trailer(s); + + if (ret < 0) + return ret; + + if (0 < mp3->xing_header.offset) + mp3_fix_xing(s); + return 0; } @@ -248,7 +471,7 @@ AVOutputFormat ff_mp3_muxer = { CODEC_ID_MP3, CODEC_ID_NONE, mp3_write_header, - ff_raw_write_packet, + mp3_write_packet, mp3_write_trailer, AVFMT_NOTIMESTAMPS, .priv_class = &mp3_muxer_class, |