【转】打包AAC码流到FLV文件

Overview

Flash Video(简称FLV),是一种流行的网络格式。目前国内外大部分视频分享网站都是采用的这种格式.

File Structure

从整个文件上开看,FLV是由The FLV header 和 The FLV File Body 组成.

1.The FLV header

Field	Type	Comment
Signature	UI8	Signature byte always 'F' (0x46)
Signature	UI8	Signature byte always 'L' (0x4C)
Signature	UI8	Signature byte always 'V' (0x56)
Version	UI8	File version (for example, 0x01 for FLV version 1)
TypeFlagsReserved	UB [5]	Shall be 0
TypeFlagsAudio	UB [1]	1 = Audio tags are present
TypeFlagsReserved	UB [1]	Shall be 0
TypeFlagsVideo	UB [1]	1 = Video tags are present
DataOffset	UI32	The length of this header in bytes

Signature: FLV 文件的前3个字节为固定的‘F’‘L’‘V’,用来标识这个文件是flv格式的.在做格式探测的时候，

如果发现前3个字节为“FLV”，就认为它是flv文件.

Version: 第4个字节表示flv版本号.

Flags: 第5个字节中的第0位和第2位,分别表示 video 与 audio 存在的情况.(1表示存在,0表示不存在)

DataOffset : 最后4个字节表示FLV header 长度.

2.The FLV File Body

Field	Type	Comment
PreviousTagSize0	UI32	Always 0
Tag1	FLVTAG	First tag
PreviousTagSize1	UI32	Size of previous tag, including its header, in bytes. For FLV version1, this value is 11 plus the DataSize of the previous tag.
Tag2	FLVTAG	Second tag
...	...	...
PreviousTagSizeN-1	UI32	Size of second-to-last tag, including its header, in bytes.
TagN	FLVTAG	Last tag
PreviousTagSizeN	UI32	Size of last tag, including its header, in bytes

FLV header之后,就是 FLV File Body.

FLV File Body是由一连串的back-pointers + tags构成.back-pointers就是4个字节数据,表示前一个tag的size.

FLV Tag Definition

FLV文件中的数据都是由一个个TAG组成,TAG里面的数据可能是video、audio、scripts.

下表是TAG的结构:

1.FLVTAG

Field	Type	Comment
Reserved	UB [2]	Reserved for FMS, should be 0
Filter	UB [1]	Indicates if packets are filtered. 0 = No pre-processing required. 1 = Pre-processing (such as decryption) of the packet is required before it can be rendered. Shall be 0 in unencrypted files, and 1 for encrypted tags. See Annex F. FLV Encryption for the use of filters.
TagType	UB [5]	Type of contents in this tag. The following types are defined: 8 = audio 9 = video 18 = script data
DataSize	UI24	Length of the message. Number of bytes after StreamID to end of tag (Equal to length of the tag – 11)
Timestamp	UI24	Time in milliseconds at which the data in this tag applies. This value is relative to the first tag in the FLV file, which always has a timestamp of 0.
TimestampExtended	UI8	Extension of the Timestamp field to form a SI32 value. This field represents the upper 8 bits, while the previous Timestamp field represents the lower 24 bits of the time in milliseconds.
StreamID	UI24	Always 0.
AudioTagHeader	IF TagType == 8 AudioTagHeader
VideoTagHeader	IF TagType == 9 VideoTagHeader
EncryptionHeader	IF Filter == 1 EncryptionTagHeader
FilterParams	IF Filter == 1 FilterParams
Data	IF TagType == 8 AUDIODATA IF TagType == 9 VIDEODATA IF TagType == 18 SCRIPTDATA	Data specific for each media type.

TagType: TAG中第1个字节中的前5位表示这个TAG中包含数据的类型,8 = audio,9 = video,18 = script data.

DataSize:StreamID之后的数据长度.

Timestamp和TimestampExtended组成了这个TAG包数据的PTS信息,记得刚开始做FVL demux的时候，并没有考虑TimestampExtended的值,直接就把Timestamp默认为是PTS，后来发生的现象就是画面有跳帧的现象,后来才仔细看了一下文档发现真正数据的PTS是PTS= Timestamp | TimestampExtended<<24.

StreamID之后的数据就是每种格式的情况不一样了，接下格式进行详细的介绍.

Audio Tags

如果TAG包中的TagType==8时，就表示这个TAG是audio。

StreamID之后的数据就表示是AudioTagHeader，AudioTagHeader结构如下：

Field	Type	Comment
SoundFormat	UB [4]	Format of SoundData. The following values are defined: 0 = Linear PCM, platform endian 1 = ADPCM 2 = MP3 3 = Linear PCM, little endian 4 = Nellymoser 16 kHz mono 5 = Nellymoser 8 kHz mono 6 = Nellymoser 7 = G.711 A-law logarithmic PCM 8 = G.711 mu-law logarithmic PCM 9 = reserved 10 = AAC 11 = Speex 14 = MP3 8 kHz 15 = Device-specific sound Formats 7, 8, 14, and 15 are reserved. AAC is supported in Flash Player 9,0,115,0 and higher. Speex is supported in Flash Player 10 and higher.
SoundRate	UB [2]	Sampling rate. The following values are defined: 0 = 5.5 kHz 1 = 11 kHz 2 = 22 kHz 3 = 44 kHz
SoundSize	UB [1]	Size of each audio sample. This parameter only pertains to uncompressed formats. Compressed formats always decode to 16 bits internally. 0 = 8-bit samples 1 = 16-bit samples
SoundType	UB [1]	Mono or stereo sound 0 = Mono sound 1 = Stereo sound
AACPacketType	IF SoundFormat == 10 UI8	The following values are defined: 0 = AAC sequence header 1 = AAC raw

AudioTagHeader的头1个字节，也就是接跟着StreamID的1个字节包含着音频类型、采样率等的基本信息.表里列的十分清楚.

AudioTagHeader之后跟着的就是AUDIODATA数据了，也就是audio payload 但是这里有个特例，如果音频格式（SoundFormat）是10 = AAC，AudioTagHeader中会多出1个字节的数据AACPacketType，这个字段来表示AACAUDIODATA的类型：0 = AAC sequence header，1 = AAC raw。

Field	Type	Comment
Data	IF AACPacketType ==0 AudioSpecificConfig	The AudioSpecificConfig is defined in ISO14496-3. Note that this is not the same as the contents of the esds box from an MP4/F4V file.
	ELSE IF AACPacketType == 1 Raw AAC frame data in UI8 [ ]	audio payload

AAC sequence header也就是包含了AudioSpecificConfig，AudioSpecificConfig包含着一些更加详细音频的信息，AudioSpecificConfig的定义在ISO14496-3中1.6.2.1 AudioSpecificConfig，这里就不详细贴了。而且在ffmpeg中有对AudioSpecificConfig解析的函数，ff_mpeg4audio_get_config(),可以对比的看一下，理解更深刻。

AAC raw 这种包含的就是音频ES流了，也就是audio payload.

在FLV的文件中，一般情况下 AAC sequence header 这种包只出现1次，而且是第一个audio tag，为什么要提到这种tag，因为当时在做FLVdemux的时候，如果是AAC的音频，需要在每帧AAC ES流前边添加7个字节ADST头,ADST在音频的格式中会详细解读，这是解码器通用的格式，就是AAC的纯ES流要打包成ADST格式的AAC文件，解码器才能正常播放.就是在打包ADST的时候，需要samplingFrequencyIndex这个信息，samplingFrequencyIndex最准确的信息是在AudioSpecificConfig中，所以就对AudioSpecificConfig进行解析并得到了samplingFrequencyIndex。

到这步你就完全可以把FLV 文件中的音频信息及数据提取出来，送给音频解码器正常播放了。

Video Tags

如果TAG包中的TagType==9时，就表示这个TAG是video.

StreamID之后的数据就表示是VideoTagHeader，VideoTagHeader结构如下：

Field	Type	Comment
Frame Type	UB [4]	Type of video frame. The following values are defined: 1 = key frame (for AVC, a seekable frame) 2 = inter frame (for AVC, a non-seekable frame) 3 = disposable inter frame (H.263 only) 4 = generated key frame (reserved for server use only) 5 = video info/command frame
CodecID	UB [4]	Codec Identifier. The following values are defined: 2 = Sorenson H.263 3 = Screen video 4 = On2 VP6 5 = On2 VP6 with alpha channel 6 = Screen video version 2 7 = AVC
AVCPacketType	IF CodecID == 7 UI8	The following values are defined: 0 = AVC sequence header 1 = AVC NALU 2 = AVC end of sequence (lower level NALU sequence ender is not required or supported)
CompositionTime	IF CodecID == 7 SI24	IF AVCPacketType == 1 Composition time offset ELSE 0 See ISO 14496-12, 8.15.3 for an explanation of composition times. The offset in an FLV file is always in milliseconds.

VideoTagHeader的头1个字节，也就是接跟着StreamID的1个字节包含着视频帧类型及视频CodecID最基本信息.表里列的十分清楚.

VideoTagHeader之后跟着的就是VIDEODATA数据了，也就是video payload.当然就像音频AAC一样，这里也有特例就是如果视频的格式是AVC（H.264）的话，VideoTagHeader会多出4个字节的信息.

AVCPacketType 和 CompositionTime。AVCPacketType 表示接下来 VIDEODATA （AVCVIDEOPACKET）的内容：

IF AVCPacketType == 0 AVCDecoderConfigurationRecord（AVC sequence header）
IF AVCPacketType == 1 One or more NALUs (Full frames are required)

AVCDecoderConfigurationRecord.包含着是H.264解码相关比较重要的sps和pps信息，再给AVC解码器送数据流之前一定要把sps和pps信息送出，否则的话解码器不能正常解码。而且在解码器stop之后再次start之前，如seek、快进快退状态切换等，都需要重新送一遍sps和pps的信息.AVCDecoderConfigurationRecord在FLV文件中一般情况也是出现1次，也就是第一个video tag.

AVCDecoderConfigurationRecord的定义在ISO 14496-15, 5.2.4.1中，这里不在详细贴，

SCRIPTDATA

如果TAG包中的TagType==18时，就表示这个TAG是SCRIPT.

SCRIPTDATA 结构十分复杂，定义了很多格式类型，每个类型对应一种结构.

Field	Type	Comment
Type	UI8	Type of the ScriptDataValue. The following types are defined: 0 = Number 1 = Boolean 2 = String 3 = Object 4 = MovieClip (reserved, not supported) 5 = Null 6 = Undefined 7 = Reference 8 = ECMA array 9 = Object end marker 10 = Strict array 11 = Date 12 = Long string
ScriptDataValue	IF Type == 0 DOUBLE IF Type == 1 UI8 IF Type == 2 SCRIPTDATASTRING IF Type == 3 SCRIPTDATAOBJECT IF Type == 7 UI16 IF Type == 8 SCRIPTDATAECMAARRAY IF Type == 10 SCRIPTDATASTRICTARRAY IF Type == 11 SCRIPTDATADATE IF Type == 12 SCRIPTDATALONGSTRING	Script data value. The Boolean value is (ScriptDataValue ≠ 0).

类型在FLV的官方文档中都有详细介绍.

onMetaData

onMetaData 是SCRIPTDATA中对我们来说十分重要的信息，结构如下表：

Property Name	Type	Comment
audiocodecid	Number	Audio codec ID used in the file (see E.4.2.1 for available SoundFormat values)
audiodatarate	Number	Audio bit rate in kilobits per second
audiodelay	Number	Delay introduced by the audio codec in seconds
audiosamplerate	Number	Frequency at which the audio stream is replayed
audiosamplesize	Number	Resolution of a single audio sample
canSeekToEnd	Boolean	Indicating the last video frame is a key frame
creationdate	String	Creation date and time
duration	Number	Total duration of the file in seconds
filesize	Number	Total size of the file in bytes
framerate	Number	Number of frames per second
height	Number	Height of the video in pixels
stereo	Boolean	Indicating stereo audio
videocodecid	Number	Video codec ID used in the file (see E.4.3.1 for available CodecID values)
videodatarate	Number	Video bit rate in kilobits per second
width	Number	Width of the video in pixels

这里面的duration、filesize、视频的width、height等这些信息对我们来说很有用.

keyframes

当时在做flv demux的时候，发现官方的文档中并没有对keyframes index做描述，但是flv的这种结构每个tag又不像TS有同步头，如果没有keyframes index 的话，seek及快进快退的效果会非常差，因为需要一个tag一个tag的顺序读取。后来通过网络查一些资料，发现了一个keyframes的信息藏在SCRIPTDATA中。

keyframes几乎是一个非官方的标准，也就是民间标准.在网上已经很难看到flv文件格式，但是metadata里面不包含 keyframes项目的视频 . 两个常用的操作metadata的工具是flvtool2和FLVMDI，都是把keyframes作为一个默认的元信息项目.在FLVMDI的主页(http://www.buraks.com/flvmdi/)上有描述：

keyframes: (Object) This object is added only if you specify the /k switch. 'keyframes' is known to FLVMDI and if /k switch is not specified, 'keyframes' object will be deleted.
'keyframes' object has 2 arrays: 'filepositions' and 'times'. Both arrays have the same number of elements, which is equal to the number of key frames in the FLV. Values in times array are in 'seconds'. Each correspond to the timestamp of the n'th key frame. Values in filepositions array are in 'bytes'. Each correspond to the fileposition of the nth key frame video tag (which starts with byte tag type 9).

也就是说keyframes中包含着2个内容 'filepositions' and 'times'分别指的是关键帧的文件位置和关键帧的PTS.通过keyframes可以建立起自己的Index，然后再seek和快进快退的操作中，快速有效的跳转到你想要找的关键帧的位置进行处理。

rtmpdump可以下载rtmp流并保存成flv文件。
如果要对流中的音频或视频单独处理，需要根据flv协议分别提取。
简单修改rtmpdump代码，增加相应功能。
1 提取音频:
rtmpdump程序在Download函数中循环下载:
....
do
{
....
nRead = RTMP_Read(rtmp, buffer, bufferSize);
....
}while(!RTMP_ctrlC && nRead > -1 && RTMP_IsConnected(rtmp) && !RTMP_IsTimedout(rtmp));
....

原程序是收到后写文件，生成flv。
现在，在写之前分别提取音视频，提取音频比较简单，直接分析buffer(参考RTMP_Write函数里的方法).
注意的是，rtmpdump里用的是RTMP_Read来接收，注意它的参数。为了方便，也可以直接用RTMP_ReadPacket。后面的视频使用RTMP_ReadPacket来接收并处理。

int RTMP_Write2(RTMP *r, const char *buf, int size)
{
RTMPPacket *pkt = &r->m_write;
char *pend, *enc;
int s2 = size, ret, num;

if (size < 11) {
/* FLV pkt too small */
return 0;
}

if (buf[0] == 'F' && buf[1] == 'L' && buf[2] == 'V')
{
buf += 13;
s2 -= 13;
}

pkt->m_packetType = *buf++;
pkt->m_nBodySize = AMF_DecodeInt24(buf);
buf += 3;
pkt->m_nTimeStamp = AMF_DecodeInt24(buf);
buf += 3;
pkt->m_nTimeStamp |= *buf++ << 24;
buf += 3;
s2 -= 11;

if (((pkt->m_packetType == RTMP_PACKET_TYPE_AUDIO
|| pkt->m_packetType == RTMP_PACKET_TYPE_VIDEO) &&
!pkt->m_nTimeStamp) || pkt->m_packetType == RTMP_PACKET_TYPE_INFO)
{
pkt->m_headerType = RTMP_PACKET_SIZE_LARGE;
if (pkt->m_packetType == RTMP_PACKET_TYPE_INFO)
  pkt->m_nBodySize += 16;
}
else
{
pkt->m_headerType = RTMP_PACKET_SIZE_MEDIUM;
}

BYTE outbuf2[640];
int nLen2 = 640;

AVManager::GetInstance()->Decode((BYTE*)(pkt->m_body+1), pkt->m_nBodySize-1, outbuf2, nLen2);
//实际音频内容为pkt->m_body+1，大小是pkt->m_nBodySize-1。这里的声音是speex编码。
为什么跳过第一字节，可以参考：http://bbs.rosoo.net/thread-16488-1-1.html

evt_OnReceivePacket((char*)outbuf2, nLen2);//回调出来

RTMPPacket_Free(pkt);
pkt->m_nBytesRead = 0;

2
视频处理
可以参考rtmpsrv.c
把nRead = RTMP_Read(rtmp, buffer, bufferSize);改成：

RTMPPacket pc = { 0 }, ps = { 0 };
bool bFirst = true;
while (RTMP_ReadPacket(rtmp, &pc))
{
if (RTMPPacket_IsReady(&pc))
{
     if (pc.m_packetType == RTMP_PACKET_TYPE_VIDEO && RTMP_ClientPacket(rtmp, &pc))
    {
        bool bIsKeyFrame = false;
if (result == 0x17)//I frame
    {
        bIsKeyFrame = true;
    }
    else if (result == 0x27)
    {
        bIsKeyFrame = false;
    }
static unsigned char const start_code[4] = {0x00, 0x00, 0x00, 0x01};
fwrite(start_code, 1, 4, pf );
//int ret = fwrite(pc.m_body + 9, 1, pc.m_nBodySize-9, pf);

if( bFirst) {

//AVCsequence header

//ioBuffer.put(foredata);

//获取sps

int spsnum = data[10]&0x1f;

int number_sps = 11;

int count_sps = 1;

while (count_sps<=spsnum){

int spslen =(data[number_sps]&0x000000FF)<<8 |(data[number_sps+1]&0x000000FF);

number_sps += 2;

fwrite(data+number_sps, 1, spslen, pf );
fwrite(start_code, 1, 4, pf );

//ioBuffer.put(data,number_sps, spslen);
//ioBuffer.put(foredata);

number_sps += spslen;

count_sps ++;

}

//获取pps

int ppsnum = data[number_sps]&0x1f;

int number_pps = number_sps+1;

int count_pps = 1;

while (count_pps<=ppsnum){

int ppslen =(data[number_pps]&0x000000FF)<<8|data[number_pps+1]&0x000000FF;

number_pps += 2;

//ioBuffer.put(data,number_pps,ppslen);

//ioBuffer.put(foredata);

fwrite(data+number_pps, 1, ppslen, pf );
fwrite(start_code, 1, 4, pf );

number_pps += ppslen;

count_pps ++;

}

bFirst =false;

} else {

//AVCNALU

int len =0;

int num =5;

//ioBuffer.put(foredata);

while(num<pc.m_nbodysize)
{

len =(data[num]&0x000000FF)<<24|(data[num+1]&0x000000FF)<<16|(data[num+2]&0x000000FF)<<8|data[num+3]&0x000000FF;

num = num+4;

//ioBuffer.put(data,num,len);

//ioBuffer.put(foredata);

fwrite(data+num, 1, len, pf );
fwrite(start_code, 1, 4, pf );

num = num + len;

}

}

}
}