=================================================================
音视频入门基础:WAV专题系列文章:
音视频入门基础:WAV专题(1)——使用FFmpeg命令生成WAV音频文件
音视频入门基础:WAV专题(2)——WAV格式简介
音视频入门基础:WAV专题(3)——FFmpeg源码中,判断某文件是否为WAV音频文件的实现
音视频入门基础:WAV专题(4)——FFmpeg源码中获取WAV文件音频压缩编码格式、采样频率、声道数量、采样位数、码率的实现
音视频入门基础:WAV专题(5)——FFmpeg源码中解码WAV Header的实现
=================================================================
一、引言
执行FFmpeg命令:
./ffmpeg -i XXX.wav
FFmpeg内部会调用wav_probe函数检测该文件是否为WAV格式的音频文件(具体可以参考:《音视频入门基础:WAV专题(3)——FFmpeg源码中,判断某文件是否为WAV音频文件的实现》)。然后如果检测出该文件为WAV格式的音频文件,会调用wav_read_header函数解码WAV Header。
二、wav_read_header函数的定义
wav_read_header函数定义在FFmpeg源码(本文演示用的FFmpeg源码版本为5.0.3)的源文件libavformat/wavdec.c中:
/* wav input */
static int wav_read_header(AVFormatContext *s)
{
int64_t size, av_uninit(data_size);
int64_t sample_count = 0;
int rf64 = 0, bw64 = 0;
uint32_t tag;
AVIOContext *pb = s->pb;
AVStream *st = NULL;
WAVDemuxContext *wav = s->priv_data;
int ret, got_fmt = 0, got_xma2 = 0;
int64_t next_tag_ofs, data_ofs = -1;
wav->unaligned = avio_tell(s->pb) & 1;
wav->smv_data_ofs = -1;
/* read chunk ID */
tag = avio_rl32(pb);
switch (tag) {
case MKTAG('R', 'I', 'F', 'F'):
break;
case MKTAG('R', 'I', 'F', 'X'):
wav->rifx = 1;
break;
case MKTAG('R', 'F', '6', '4'):
rf64 = 1;
break;
case MKTAG('B', 'W', '6', '4'):
bw64 = 1;
break;
default:
av_log(s, AV_LOG_ERROR, "invalid start code %s in RIFF header\n",
av_fourcc2str(tag));
return AVERROR_INVALIDDATA;
}
/* read chunk size */
avio_rl32(pb);
/* read format */
if (avio_rl32(pb) != MKTAG('W', 'A', 'V', 'E')) {
av_log(s, AV_LOG_ERROR, "invalid format in RIFF header\n");
return AVERROR_INVALIDDATA;
}
if (rf64 || bw64) {
if (avio_rl32(pb) != MKTAG('d', 's', '6', '4'))
return AVERROR_INVALIDDATA;
size = avio_rl32(pb);
if (size < 24)
return AVERROR_INVALIDDATA;
avio_rl64(pb); /* RIFF size */
data_size = avio_rl64(pb);
sample_count = avio_rl64(pb);
if (data_size < 0 || sample_count < 0) {
av_log(s, AV_LOG_ERROR, "negative data_size and/or sample_count in "
"ds64: data_size = %"PRId64", sample_count = %"PRId64"\n",
data_size, sample_count);
return AVERROR_INVALIDDATA;
}
avio_skip(pb, size - 24); /* skip rest of ds64 chunk */
}
/* Create the audio stream now so that its index is always zero */
st = avformat_new_stream(s, NULL);
if (!st)
return AVERROR(ENOMEM);
for (;;) {
AVStream *vst;
size = next_tag(pb, &tag, wav->rifx);
next_tag_ofs = avio_tell(pb) + size;
if (avio_feof(pb))
break;
switch (tag) {
case MKTAG('f', 'm', 't', ' '):
/* only parse the first 'fmt ' tag found */
if (!got_xma2 && !got_fmt && (ret = wav_parse_fmt_tag(s, size, st)) < 0) {
return ret;
} else if (got_fmt)
av_log(s, AV_LOG_WARNING, "found more than one 'fmt ' tag\n");
got_fmt = 1;
break;
case MKTAG('X', 'M', 'A', '2'):
/* only parse the first 'XMA2' tag found */
if (!got_fmt && !got_xma2 && (ret = wav_parse_xma2_tag(s, size, st)) < 0) {
return ret;
} else if (got_xma2)
av_log(s, AV_LOG_WARNING, "found more than one 'XMA2' tag\n");
got_xma2 = 1;
break;
case MKTAG('d', 'a', 't', 'a'):
if (!(pb->seekable & AVIO_SEEKABLE_NORMAL) && !got_fmt && !got_xma2) {
av_log(s, AV_LOG_ERROR,
"found no 'fmt ' tag before the 'data' tag\n");
return AVERROR_INVALIDDATA;
}
if (rf64 || bw64) {
next_tag_ofs = wav->data_end = avio_tell(pb) + data_size;
} else if (size != 0xFFFFFFFF) {
data_size = size;
next_tag_ofs = wav->data_end = size ? next_tag_ofs : INT64_MAX;
} else {
av_log(s, AV_LOG_WARNING, "Ignoring maximum wav data size, "
"file may be invalid\n");
data_size = 0;
next_tag_ofs = wav->data_end = INT64_MAX;
}
data_ofs = avio_tell(pb);
/* don't look for footer metadata if we can't seek or if we don't
* know where the data tag ends
*/
if (!(pb->seekable & AVIO_SEEKABLE_NORMAL) || (!(rf64 && !bw64) && !size))
goto break_loop;
break;
case MKTAG('f', 'a', 'c', 't'):
if (!sample_count)
sample_count = (!wav->rifx ? avio_rl32(pb) : avio_rb32(pb));
break;
case MKTAG('b', 'e', 'x', 't'):
if ((ret = wav_parse_bext_tag(s, size)) < 0)
return ret;
break;
case MKTAG('S','M','V','0'):
if (!got_fmt) {
av_log(s, AV_LOG_ERROR, "found no 'fmt ' tag before the 'SMV0' tag\n");
return AVERROR_INVALIDDATA;
}
// SMV file, a wav file with video appended.
if (size != MKTAG('0','2','0','0')) {
av_log(s, AV_LOG_ERROR, "Unknown SMV version found\n");
goto break_loop;
}
av_log(s, AV_LOG_DEBUG, "Found SMV data\n");
wav->smv_given_first = 0;
vst = avformat_new_stream(s, NULL);
if (!vst)
return AVERROR(ENOMEM);
wav->vst = vst;
avio_r8(pb);
vst->id = 1;
vst->codecpar->codec_type = AVMEDIA_TYPE_VIDEO;
vst->codecpar->codec_id = AV_CODEC_ID_SMVJPEG;
vst->codecpar->width = avio_rl24(pb);
vst->codecpar->height = avio_rl24(pb);
if ((ret = ff_alloc_extradata(vst->codecpar, 4)) < 0) {
av_log(s, AV_LOG_ERROR, "Could not allocate extradata.\n");
return ret;
}
size = avio_rl24(pb);
wav->smv_data_ofs = avio_tell(pb) + (size - 5) * 3;
avio_rl24(pb);
wav->smv_block_size = avio_rl24(pb);
if (!wav->smv_block_size)
return AVERROR_INVALIDDATA;
avpriv_set_pts_info(vst, 32, 1, avio_rl24(pb));
vst->duration = avio_rl24(pb);
avio_rl24(pb);
avio_rl24(pb);
wav->smv_frames_per_jpeg = avio_rl24(pb);
if (wav->smv_frames_per_jpeg > 65536) {
av_log(s, AV_LOG_ERROR, "too many frames per jpeg\n");
return AVERROR_INVALIDDATA;
}
AV_WL32(vst->codecpar->extradata, wav->smv_frames_per_jpeg);
goto break_loop;
case MKTAG('L', 'I', 'S', 'T'):
case MKTAG('l', 'i', 's', 't'):
if (size < 4) {
av_log(s, AV_LOG_ERROR, "too short LIST tag\n");
return AVERROR_INVALIDDATA;
}
switch (avio_rl32(pb)) {
case MKTAG('I', 'N', 'F', 'O'):
ff_read_riff_info(s, size - 4);
break;
case MKTAG('a', 'd', 't', 'l'):
if (s->nb_chapters > 0) {
while (avio_tell(pb) < next_tag_ofs &&
!avio_feof(pb)) {
char cue_label[512];
unsigned id, sub_size;
if (avio_rl32(pb) != MKTAG('l', 'a', 'b', 'l'))
break;
sub_size = avio_rl32(pb);
if (sub_size < 5)
break;
id = avio_rl32(pb);
avio_get_str(pb, sub_size - 4, cue_label, sizeof(cue_label));
avio_skip(pb, avio_tell(pb) & 1);
for (int i = 0; i < s->nb_chapters; i++) {
if (s->chapters[i]->id == id) {
av_dict_set(&s->chapters[i]->metadata, "title", cue_label, 0);
break;
}
}
}
}
break;
}
break;
case MKTAG('I', 'D', '3', ' '):
case MKTAG('i', 'd', '3', ' '): {
ID3v2ExtraMeta *id3v2_extra_meta;
ff_id3v2_read_dict(pb, &ffformatcontext(s)->id3v2_meta, ID3v2_DEFAULT_MAGIC, &id3v2_extra_meta);
if (id3v2_extra_meta) {
ff_id3v2_parse_apic(s, id3v2_extra_meta);
ff_id3v2_parse_chapters(s, id3v2_extra_meta);
ff_id3v2_parse_priv(s, id3v2_extra_meta);
}
ff_id3v2_free_extra_meta(&id3v2_extra_meta);
}
break;
case MKTAG('c', 'u', 'e', ' '):
if (size >= 4 && got_fmt && st->codecpar->sample_rate > 0) {
AVRational tb = {1, st->codecpar->sample_rate};
unsigned nb_cues = avio_rl32(pb);
if (size >= nb_cues * 24LL + 4LL) {
for (int i = 0; i < nb_cues; i++) {
unsigned offset, id = avio_rl32(pb);
if (avio_feof(pb))
return AVERROR_INVALIDDATA;
avio_skip(pb, 16);
offset = avio_rl32(pb);
if (!avpriv_new_chapter(s, id, tb, offset, AV_NOPTS_VALUE, NULL))
return AVERROR(ENOMEM);
}
}
}
break;
}
/* seek to next tag unless we know that we'll run into EOF */
if ((avio_size(pb) > 0 && next_tag_ofs >= avio_size(pb)) ||
wav_seek_tag(wav, pb, next_tag_ofs, SEEK_SET) < 0) {
break;
}
}
break_loop:
if (!got_fmt && !got_xma2) {
av_log(s, AV_LOG_ERROR, "no 'fmt ' or 'XMA2' tag found\n");
return AVERROR_INVALIDDATA;
}
if (data_ofs < 0) {
av_log(s, AV_LOG_ERROR, "no 'data' tag found\n");
return AVERROR_INVALIDDATA;
}
avio_seek(pb, data_ofs, SEEK_SET);
if (data_size > (INT64_MAX>>3)) {
av_log(s, AV_LOG_WARNING, "Data size %"PRId64" is too large\n", data_size);
data_size = 0;
}
if ( st->codecpar->bit_rate > 0 && data_size > 0
&& st->codecpar->sample_rate > 0
&& sample_count > 0 && st->codecpar->channels > 1
&& sample_count % st->codecpar->channels == 0) {
if (fabs(8.0 * data_size * st->codecpar->channels * st->codecpar->sample_rate /
sample_count /st->codecpar->bit_rate - 1.0) < 0.3)
sample_count /= st->codecpar->channels;
}
if ( data_size > 0 && sample_count && st->codecpar->channels
&& (data_size << 3) / sample_count / st->codecpar->channels > st->codecpar->bits_per_coded_sample + 1) {
av_log(s, AV_LOG_WARNING, "ignoring wrong sample_count %"PRId64"\n", sample_count);
sample_count = 0;
}
/* G.729 hack (for Ticket4577)
* FIXME: Come up with cleaner, more general solution */
if (st->codecpar->codec_id == AV_CODEC_ID_G729 && sample_count && (data_size << 3) > sample_count) {
av_log(s, AV_LOG_WARNING, "ignoring wrong sample_count %"PRId64"\n", sample_count);
sample_count = 0;
}
if (!sample_count || av_get_exact_bits_per_sample(st->codecpar->codec_id) > 0)
if ( st->codecpar->channels
&& data_size
&& av_get_bits_per_sample(st->codecpar->codec_id)
&& wav->data_end <= avio_size(pb))
sample_count = (data_size << 3)
/
(st->codecpar->channels * (uint64_t)av_get_bits_per_sample(st->codecpar->codec_id));
if (sample_count)
st->duration = sample_count;
if (st->codecpar->codec_id == AV_CODEC_ID_PCM_S32LE &&
st->codecpar->block_align == st->codecpar->channels * 4 &&
st->codecpar->bits_per_coded_sample == 32 &&
st->codecpar->extradata_size == 2 &&
AV_RL16(st->codecpar->extradata) == 1) {
st->codecpar->codec_id = AV_CODEC_ID_PCM_F16LE;
st->codecpar->bits_per_coded_sample = 16;
} else if (st->codecpar->codec_id == AV_CODEC_ID_PCM_S24LE &&
st->codecpar->block_align == st->codecpar->channels * 4 &&
st->codecpar->bits_per_coded_sample == 24) {
st->codecpar->codec_id = AV_CODEC_ID_PCM_F24LE;
} else if (st->codecpar->codec_id == AV_CODEC_ID_XMA1 ||
st->codecpar->codec_id == AV_CODEC_ID_XMA2) {
st->codecpar->block_align = 2048;
} else if (st->codecpar->codec_id == AV_CODEC_ID_ADPCM_MS && st->codecpar->channels > 2 &&
st->codecpar->block_align < INT_MAX / st->codecpar->channels) {
st->codecpar->block_align *= st->codecpar->channels;
}
ff_metadata_conv_ctx(s, NULL, wav_metadata_conv);
ff_metadata_conv_ctx(s, NULL, ff_riff_info_conv);
set_spdif(s, wav);
return 0;
}
形参s:既是输入型参数也是输出型参数,指向AVFormatContext类型的变量。s->pb包含整个WAV Header的二进制数据。执行wav_read_header函数后,(WAVDemuxContext *)(s->priv_data)->vst->codecpar中的成员变量会被赋值为WAV Header中Format chunk中的信息(包括音频压缩编码格式、采样频率、声道数量、采样位数、码率)。
返回值:返回0表示解码WAV Header成功。返回一个负数表示解码WAV Header失败。
三、wav_read_header函数的内部实现分析
(一)读取WAV Header中的“区块编号”
WAV Header中的第一个区块为“区块编号”,通过下面语句读取“区块编号”的内容(关于avio_rXXX系列函数的用法可以参考:《FFmpeg源码:avio_r8、avio_rl16、avio_rl24、avio_rl32、avio_rl64函数分析》):
/* read chunk ID */
tag = avio_rl32(pb);
有的同学看到这里可能会有疑问:根据文章《音视频入门基础:WAV专题(2)——WAV格式简介》的描述,“区块编号”不是大端字节序的吗?而avio_rl32函数是按照小端模式读取四个字节数据。所以为什么是用avio_rl32函数而不是avio_rb32函数读取呢?avio_rb32函数才是按照大端模式读取四个字节数据的吧?
原因是这样的,在语句tag = avio_rl32(pb)之后,又通过MKTAG函数(关于MKTAG和MKBETAG宏定义的用法可以参考:《FFmpeg源码:MKTAG和MKBETAG宏定义分析》)将字符(比如'R', 'I', 'F', 'F')转换为整形,按小端模式存贮,让其跟tag进行比较。所以这就是所谓的“负负得正”,“用avio_rl32函数读取然后跟MKTAG转换出来的整数进行比较”,这个跟“用avio_rb32函数读取然后用MKBETAG转换出来的整数进行比较”,效果是一样的:
switch (tag) {
case MKTAG('R', 'I', 'F', 'F'):
break;
case MKTAG('R', 'I', 'F', 'X'):
wav->rifx = 1;
break;
case MKTAG('R', 'F', '6', '4'):
rf64 = 1;
break;
case MKTAG('B', 'W', '6', '4'):
bw64 = 1;
break;
default:
av_log(s, AV_LOG_ERROR, "invalid start code %s in RIFF header\n",
av_fourcc2str(tag));
return AVERROR_INVALIDDATA;
}
所以下面代码块:
/* read chunk ID */
tag = avio_rl32(pb);
switch (tag) {
case MKTAG('R', 'I', 'F', 'F'):
break;
case MKTAG('R', 'I', 'F', 'X'):
wav->rifx = 1;
break;
case MKTAG('R', 'F', '6', '4'):
rf64 = 1;
break;
case MKTAG('B', 'W', '6', '4'):
bw64 = 1;
break;
default:
av_log(s, AV_LOG_ERROR, "invalid start code %s in RIFF header\n",
av_fourcc2str(tag));
return AVERROR_INVALIDDATA;
}
等价于:
/* read chunk ID */
tag = avio_rb32(pb);
switch (tag) {
case MKBETAG('R', 'I', 'F', 'F'):
break;
case MKBETAG('R', 'I', 'F', 'X'):
wav->rifx = 1;
break;
case MKBETAG('R', 'F', '6', '4'):
rf64 = 1;
break;
case MKBETAG('B', 'W', '6', '4'):
bw64 = 1;
break;
default:
av_log(s, AV_LOG_ERROR, "invalid start code %s in RIFF header\n",
av_fourcc2str(tag));
return AVERROR_INVALIDDATA;
}
读取到“区块编号”后,如果值等于“RIFF”,表示该文件遵守RIFF格式的规则,按默认处理;如果值等于“RIFX”,让wav->rifx赋值为1,表示该文件遵守RIFX格式的规则;如果值等于“RF64”,让变量rf64赋值为1,表示这是WAVE 64位扩展格式中的一种:WAV RF64;如果值等于“BW64”,让变量bw64赋值为1,表示这是WAVE 64位扩展格式中的一种:WAV BW64;如果值不为上述,打印日志:"invalid start code XXX in RIFF header"并返回AVERROR_INVALIDDATA表示WAV Header中的数据不合法。
(二)读取WAV Header中的“总区块大小”
WAV Header中的第二个区块为“总区块大小”,通过下面语句读取“总区块大小”。由于该值可以由WAV Header中的其它值推导出来,所以对于FFmpeg来讲它没有意义,故FFmpeg没有将其存贮到内部的成员变量中:
/* read chunk size */
avio_rl32(pb);
(三)读取WAV Header中的“档案格式”
WAV Header中的第三个区块为“档案格式”。通过下面语句读取该值并进行比较,如果值不为“WAVE”,日志打印"invalid format in RIFF header"表示WAV Header中的数据不合法:
/* read format */
if (avio_rl32(pb) != MKTAG('W', 'A', 'V', 'E')) {
av_log(s, AV_LOG_ERROR, "invalid format in RIFF header\n");
return AVERROR_INVALIDDATA;
}
(四)WAVE 64位扩展格式处理
FFmpeg源码内部将WAV和WAVE 64位扩展格式都放到一起处理,下面代码处理文件格式为WAVE 64位时的情况:
if (rf64 || bw64) {
if (avio_rl32(pb) != MKTAG('d', 's', '6', '4'))
return AVERROR_INVALIDDATA;
size = avio_rl32(pb);
if (size < 24)
return AVERROR_INVALIDDATA;
avio_rl64(pb); /* RIFF size */
data_size = avio_rl64(pb);
sample_count = avio_rl64(pb);
if (data_size < 0 || sample_count < 0) {
av_log(s, AV_LOG_ERROR, "negative data_size and/or sample_count in "
"ds64: data_size = %"PRId64", sample_count = %"PRId64"\n",
data_size, sample_count);
return AVERROR_INVALIDDATA;
}
avio_skip(pb, size - 24); /* skip rest of ds64 chunk */
}
(五)读取WAV Header中的子区块
WAV Header中包含“Format chunk”、“Data chunk”这种必须存在的子区块,也包含Fact chunk、Cue points chunk、Playlist chunk、Associated data list chunk等可选区块。通过下面语句中的for循环和switch case语句来判断是哪种子区块,然后循环处理:
for (;;) {
AVStream *vst;
size = next_tag(pb, &tag, wav->rifx);
next_tag_ofs = avio_tell(pb) + size;
if (avio_feof(pb))
break;
switch (tag) {
case MKTAG('f', 'm', 't', ' '):
//...
break;
case MKTAG('X', 'M', 'A', '2'):
//...
break;
case MKTAG('d', 'a', 't', 'a'):
//...
break;
case MKTAG('f', 'a', 'c', 't'):
//...
break;
case MKTAG('b', 'e', 'x', 't'):
//...
break;
case MKTAG('S','M','V','0'):
//...
goto break_loop;
case MKTAG('L', 'I', 'S', 'T'):
case MKTAG('l', 'i', 's', 't'):
//...
break;
case MKTAG('I', 'D', '3', ' '):
case MKTAG('i', 'd', '3', ' '): {
//...
break;
case MKTAG('c', 'u', 'e', ' '):
//...
break;
}
}
}
(六)读取WAV Header中“Format chunk”子区块的信息
下面以读取“Format chunk”子区块中的信息为例子进行讲解如何解码子区块。ff_get_wav_header函数中通过wav_parse_fmt_tag函数解析“Format chunk”子区块:
int ff_get_wav_header(AVFormatContext *s, AVIOContext *pb,
AVCodecParameters *par, int size, int big_endian)
{
//...
for (;;) {
switch (tag) {
case MKTAG('f', 'm', 't', ' '):
/* only parse the first 'fmt ' tag found */
if (!got_xma2 && !got_fmt && (ret = wav_parse_fmt_tag(s, size, st)) < 0) {
return ret;
} else if (got_fmt)
av_log(s, AV_LOG_WARNING, "found more than one 'fmt ' tag\n");
got_fmt = 1;
break;
}
//...
}
}
而wav_parse_fmt_tag函数内部又调用了ff_get_wav_header函数来获取获取音频压缩编码格式、采样频率、声道数量、采样位数、码率信息(具体可以参考:《音视频入门基础:WAV专题(4)——FFmpeg源码中获取WAV文件音频压缩编码格式、采样频率、声道数量、采样位数、码率的实现》):
static int wav_parse_fmt_tag(AVFormatContext *s, int64_t size, AVStream *st)
{
AVIOContext *pb = s->pb;
WAVDemuxContext *wav = s->priv_data;
int ret;
/* parse fmt header */
ret = ff_get_wav_header(s, pb, st->codecpar, size, wav->rifx);
if (ret < 0)
return ret;
handle_stream_probing(st);
ffstream(st)->need_parsing = AVSTREAM_PARSE_FULL_RAW;
avpriv_set_pts_info(st, 64, 1, st->codecpar->sample_rate);
return 0;
}