基于libmad库的MP3解码简析

 MAD （libmad）是一个开源的高精度 MPEG 音频解码库，支持 MPEG-1（Layer I, Layer II 和 LayerIII（也就是 MP3）。LIBMAD 提供 24-bit 的 PCM 输出，完全是定点计算，非常适合没有浮点支持的平台上使用。使用 libmad 提供的一系列 API，就可以非常简单地实现 MP3 数据解码工作。在 libmad 的源代码文件目录下的 mad.h 文件中，可以看到绝大部分该库的数据结构和 API 等。

     网上有很多关于libmad的使用实例，在他们的基础上，我稍加总结、整理和衍生，文末给出相关参考链接，表示感谢！

 

     一、libmad库源码

 

     可以去相关网站下载，给出链接：

     http://download.chinaunix.net/download.php?id=11891&ResourceID=5910

可以根据不同的平台自行编译或者移植，略述。

 

     二、相关数据结构及函数接口简介

 

     1、struct mad_decode

struct mad_decoder {
  enum mad_decoder_mode mode;

  int options;

  struct {
    long pid;
    int in;
    int out;
  } async;

  struct {
    struct mad_stream stream;
    struct mad_frame frame;
    struct mad_synth synth;
  } *sync;

  void *cb_data;

  enum mad_flow (*input_func)(void *, struct
 mad_stream *);
  enum mad_flow (*header_func)(void *, struct
 mad_header const *);
  enum mad_flow (*filter_func)(void *,
             struct mad_stream const *, struct mad_frame *);
  enum mad_flow (*output_func)(void *,
             struct mad_header const *, struct mad_pcm *);
  enum mad_flow (*error_func)(void *, struct
 mad_stream *, struct mad_frame *);
  enum mad_flow (*message_func)(void *, void *, unsigned int *);
};

     2、struct mad_stream

struct mad_stream {
  unsigned char const *buffer;        /* input
 bitstream buffer */
  unsigned char const *bufend;        /* end of
 buffer */
  unsigned long skiplen;              /* bytes to skip
 before next frame */

  int sync;                           /* stream
 sync found */
  unsigned long freerate;             /* free bitrate (fixed) */

  unsigned char const *this_frame;    /* start
 of current frame */
  unsigned char const *next_frame;    /* start
 of next frame */
  struct mad_bitptr ptr;              /* current processing bit pointer */

  struct mad_bitptr anc_ptr;          /* ancillary bits pointer */
  unsigned int anc_bitlen;            /* number
 of ancillary bits */

  unsigned char (*main_data)[MAD_BUFFER_MDLEN];
                                      /* Layer III main_data() */
  unsigned int md_len;                /* bytes in main_data */

  int options;                        /* decoding
 options (see below) */
  enum mad_error error;               /* error code (see
 above) */
};

     三、MP3解码流程简介

 

     MP3解码有同步方式和异步方式两种，libmad是以桢为单位对MP3进行解码的，所谓同步方式是指解码函数在解码完一帧后才返回并带回出错信息，异步方式是指解码函数在调用后立即返回，通过消息传递解码状态信息。

     1、首先创建一个解码器 struct mad_decoder decoder，紧接着调用函数  mad_decoder_init(...)函数，给出这个函数的原型及定义

/*
 * NAME:    decoder->init()
 * DESCRIPTION:    initialize a decoder object with callback routines
 */
void mad_decoder_init(struct mad_decoder *decoder, void *data,
         enum mad_flow (*input_func)(void *,
                         struct mad_stream *),
         enum mad_flow (*header_func)(void *,
                         struct mad_header const *),
         enum mad_flow (*filter_func)(void *,
                         struct mad_stream const *,
                         struct mad_frame *),
         enum mad_flow (*output_func)(void *,
                         struct mad_header const *,
                         struct mad_pcm *),
         enum mad_flow (*error_func)(void *,
                         struct mad_stream *,
                         struct mad_frame *),
         enum mad_flow (*message_func)(void *,
                         void *, unsigned int *))
{
  decoder->mode = -1;

  decoder->options = 0;

  decoder->async.pid = 0;
  decoder->async.in = -1;
  decoder->async.out = -1;

  decoder->sync = 0;

  decoder->cb_data = data;

  decoder->input_func = input_func;
  decoder->header_func = header_func;
  decoder->filter_func = filter_func;
  decoder->output_func = output_func;
  decoder->error_func = error_func;
  decoder->message_func = message_func;
}

     用户编程可以用如下方式调用，可以看到从第三个参数开始，其实都是一些列的函数指针，这里初始化的目的其实是给创建的decoder注册下面即将要自己实现的这些函数。Libmad库会在解码过程中回调这些函数：

mad_decoder_init(&decoder, &buffer,
         input, 0 /* header */, 0 /* filter */, output,
         error, 0 /* message */);

     第一个参数，就是定义的解码器decoder；

     第二个参数，是一个void型的函数指针，这里也就是给你的用户空间定义私有的数据结构体用的，下面会给出具体的例子来说明其用法；

     第三个参数，input_func函数，这个是用来读取你的mp3资源的函数；

     第四个参数，header_func函数，这个顾名思义是处理mp3头部信息的函数，可以根据需要取舍；

     第五个参数，filter_func函数，也没有深入理解过，可以不必实现；
     第六个参数，output_func函数，这个是用来将解码之后的数据写入输出缓冲区或者音频设备节点的；

     第七个参数，error_func函数，是用来打印返回的解码出错信息的；

     第八个参数，message_func可以不必实现。

     2、调用mad_decoder_run(&decoder, MAD_DECODER_MODE_SYNC)函数启动解码，查看Libmad库源码可知，这个函数里面会注册一个函数指针

/*
 * NAME:    decoder->run()
 * DESCRIPTION:    run the decoder thread either synchronously or asynchronously
 */
int mad_decoder_run(struct mad_decoder *decoder, enum
 mad_decoder_mode mode)
{
  int result;
  int (*run)(struct
 mad_decoder *) = 0;

  switch (decoder->mode = mode) {
  case MAD_DECODER_MODE_SYNC:
    run = run_sync;
    break;

  case MAD_DECODER_MODE_ASYNC:
# if defined(USE_ASYNC)
    run = run_async;
# endif
    break;
  }

  if (run == 0)
    return -1;

  decoder->sync = malloc(sizeof(*decoder->sync));
  if (decoder->sync == 0)
    return -1;

  result = run(decoder);

  free(decoder->sync);
  decoder->sync = 0;

  return result;
}

     而在这个run_sync(struct mad_decoder *decoder)函数中则有一个大的while循环来依次调用

decoder->input_func(decoder->cb_data, stream)获取mp3源文件，然后交由相关库函数解码。

而后会有decoder->output_func(decoder->cb_data,  &frame->header, &synth->pcm)函数来输出解码后的数据。

     3、最后调用mad_decoder_finish(&decoder)结束解码，释放decoder资源。

     4、在input_func函数中，会调用一个很重要的函数

mad_stream_buffer(stream, buffer->start, buffer->length) ，第一个参数指向一个mad_stream变量，mad_stream结构定义在stream.h头文件里，用于记录文件的地址和当前处理的位置。第二、三个参数分别是mp3文件在内存中映像的起始地址和文件长度。mad_stream_buffer()函数将mp3文件与mad_stream结构进行关联。

 

     四、MP3解码编程实例

#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <unistd.h>
#include <sys/stat.h>
#include <sys/mman.h>
#include <fcntl.h>
#include <sys/types.h>
#include <sys/ioctl.h>
#include <sys/soundcard.h>
#include "mad.h"

#define BUFSIZE 8192

/*
 * This is a private message structure. A
 generic pointer to this structure
 * is passed to each of the callback
 functions. Put here any data you need
 * to access from within the callbacks.
 */
struct buffer {
    FILE *fp; /*file pointer*/
    unsigned int flen; /*file
 length*/
    unsigned int fpos; /*current
 position*/
    unsigned char fbuf[BUFSIZE]; /*buffer*/
    unsigned int fbsize; /*indeed
 size of buffer*/
};
typedef struct buffer mp3_file;

int soundfd; /*soundcard file*/
unsigned int prerate = 0; /*the
 pre simple rate*/

int writedsp(int c)
{
    return write(soundfd, (char *)&c, 1);
}

void set_dsp()
{
#if 0
    int format = AFMT_S16_LE;
    int channels = 2;
    int rate = 44100;

    soundfd = open("/dev/dsp", O_WRONLY);
    ioctl(soundfd, SNDCTL_DSP_SPEED,&rate);
    ioctl(soundfd, SNDCTL_DSP_SETFMT, &format);
    ioctl(soundfd, SNDCTL_DSP_CHANNELS, &channels);
#else
    if((soundfd = open("test.bin" , O_WRONLY | O_CREAT)) < 0)
    {
        fprintf(stderr , "can't open sound device!\n");
        exit(-1);
    }
#endif
}

/*
 * This is perhaps the simplest example use of the MAD high-level API.
 * Standard input is mapped into memory via mmap(), then the
 high-level API
 * is invoked with three callbacks: input, output, and error. The
 output
 * callback converts MAD's high-resolution PCM samples to 16
 bits, then
 * writes them to standard output in little-endian, stereo-interleaved
 * format.
 */

static int decode(mp3_file *mp3fp);

int main(int argc, char *argv[])
{
    long flen, fsta, fend;
    int dlen;
    mp3_file *mp3fp;

    if (argc != 2)
        return 1;

    mp3fp = (mp3_file *)malloc(sizeof(mp3_file));
    if((mp3fp->fp = fopen(argv[1], "r")) == NULL)
    {
        printf("can't open source file.\n");
        return 2;
    }
    fsta = ftell(mp3fp->fp);
    fseek(mp3fp->fp, 0, SEEK_END);
    fend = ftell(mp3fp->fp);
    flen = fend - fsta;
    if(flen > 0)
        fseek(mp3fp->fp, 0, SEEK_SET);
    fread(mp3fp->fbuf, 1, BUFSIZE, mp3fp->fp);
    mp3fp->fbsize = BUFSIZE;
    mp3fp->fpos = BUFSIZE;
    mp3fp->flen = flen;

    set_dsp();

    decode(mp3fp);

    close(soundfd);
    fclose(mp3fp->fp);

    return 0;
}

static enum mad_flow input(void *data, struct mad_stream *stream)
{
    mp3_file *mp3fp;
    int ret_code;
    int unproc_data_size; /*the
 unprocessed data's size*/
    int copy_size;

    mp3fp = (mp3_file *)data;
    if(mp3fp->fpos < mp3fp->flen) {
        unproc_data_size = stream->bufend - stream->next_frame;
        //printf("%d,
 %d, %d\n", unproc_data_size, mp3fp->fpos, mp3fp->fbsize);
        memcpy(mp3fp->fbuf, mp3fp->fbuf + mp3fp->fbsize - unproc_data_size, unproc_data_size);
        copy_size = BUFSIZE - unproc_data_size;
        if(mp3fp->fpos + copy_size > mp3fp->flen) {
            copy_size = mp3fp->flen - mp3fp->fpos;
        }
        fread(mp3fp->fbuf+unproc_data_size, 1, copy_size, mp3fp->fp);
        mp3fp->fbsize = unproc_data_size + copy_size;
        mp3fp->fpos += copy_size;

        /*Hand off the buffer to the
 mp3 input stream*/
        mad_stream_buffer(stream, mp3fp->fbuf, mp3fp->fbsize);
        ret_code = MAD_FLOW_CONTINUE;
    } else {
        ret_code = MAD_FLOW_STOP;
    }

    return ret_code;

}

/*
 * The following utility routine performs simple rounding, clipping, and
 * scaling of MAD's high-resolution samples down to 16
 bits. It does not
 * perform any dithering or noise shaping, which would be recommended to
 * obtain any exceptional audio quality. It is therefore not recommended to
 * use this routine if high-quality output is desired.
 */

static inline signed int scale(mad_fixed_t sample)
{
    /* round */
    sample += (1L << (MAD_F_FRACBITS - 16));

    /* clip */
    if (sample >= MAD_F_ONE)
        sample = MAD_F_ONE - 1;
    else if (sample < -MAD_F_ONE)
        sample = -MAD_F_ONE;

    /* quantize */
    return sample >> (MAD_F_FRACBITS + 1 - 16);
}

/*
 * This is the output callback function. It is called
 after each frame of
 * MPEG audio data has been completely decoded. The purpose of this callback
 * is to output (or play) the
 decoded PCM audio.
 */

//输出函数做相应的修改，目的是解决播放音乐时声音卡的问题。
static enum mad_flow output(void *data, struct mad_header const *header,
        struct mad_pcm *pcm)
{
    unsigned int nchannels, nsamples;
    mad_fixed_t const *left_ch, *right_ch;
    // pcm->samplerate contains
 the sampling frequency
    nchannels = pcm->channels;
    nsamples = pcm->length;
    left_ch = pcm->samples[0];
    right_ch = pcm->samples[1];
    short buf[nsamples *2];
    int i = 0;
    //printf(">>%d\n", nsamples);
    while (nsamples--) {
        signed int sample;
        // output sample(s) in 16-bit
 signed little-endian PCM
        sample = scale(*left_ch++);
        buf[i++] = sample & 0xFFFF;
        if (nchannels == 2) {
            sample = scale(*right_ch++);
            buf[i++] = sample & 0xFFFF;
        }
    }
    //fprintf(stderr, ".");
    write(soundfd, &buf[0], i * 2);
    return MAD_FLOW_CONTINUE;
}

/*
 * This is the error callback function. It is called
 whenever a decoding
 * error occurs. The error is indicated
 by stream->error; the list of
 * possible MAD_ERROR_* errors can be found in the mad.h (or stream.h)
 * header file.
 */

static enum mad_flow error(void *data,
        struct mad_stream *stream,
        struct mad_frame *frame)
{
    mp3_file *mp3fp = data;

    fprintf(stderr, "decoding error 0x%04x (%s) at byte offset %u\n",
            stream->error, mad_stream_errorstr(stream),
            stream->this_frame - mp3fp->fbuf);

    /* return MAD_FLOW_BREAK here to stop decoding (and propagate
 an error) */

    return MAD_FLOW_CONTINUE;
}

/*
 * This is the function called by main() above to perform
 all the decoding.
 * It instantiates a decoder object and configures it with the input,
 * output, and error callback
 functions above. A single call to
 * mad_decoder_run() continues until a
 callback function returns
 * MAD_FLOW_STOP (to stop decoding) or MAD_FLOW_BREAK (to stop
 decoding and
 * signal an error).
 */

static int decode(mp3_file *mp3fp)
{
    struct mad_decoder decoder;
    int result;

    /* configure input, output, and error functions */
    mad_decoder_init(&decoder, mp3fp,
            input, 0 /* header */, 0 /* filter */, output,
            error, 0 /* message */);

    /* start decoding */
    result = mad_decoder_run(&decoder, MAD_DECODER_MODE_SYNC);

    /* release the decoder */
    mad_decoder_finish(&decoder);

    return result;
}

说明：1、实例原本是基于音频OSS框架的，当然，在嵌入式领域，ALSA也是兼容OSS接口的；

        2、为了在ubuntu上调试方便，并没有直接往音频接口，而是创建了一个文件，直接往文件里面写；

        3、上述代码中的红色区域重点讲解一下

static enum mad_flow input(void *data, struct
 mad_stream *stream)
{
    mp3_file *mp3fp;
    int ret_code;
    int unproc_data_size; /*the
 unprocessed data's size*/
    int copy_size;

    mp3fp = (mp3_file *)data;
    if(mp3fp->fpos < mp3fp->flen) {
        unproc_data_size = stream->bufend - stream->next_frame;
        //printf("%d, %d, %d\n", unproc_data_size, mp3fp->fpos, mp3fp->fbsize);
        memcpy(mp3fp->fbuf, mp3fp->fbuf + mp3fp->fbsize - unproc_data_size, unproc_data_size);
        copy_size = BUFSIZE - unproc_data_size;
        if(mp3fp->fpos + copy_size > mp3fp->flen) {
            copy_size = mp3fp->flen - mp3fp->fpos;
        }
        fread(mp3fp->fbuf+unproc_data_size, 1, copy_size, mp3fp->fp);
        mp3fp->fbsize = unproc_data_size + copy_size;
        mp3fp->fpos += copy_size;

        /*Hand off the buffer to the mp3 input stream*/
        mad_stream_buffer(stream, mp3fp->fbuf, mp3fp->fbsize);
        ret_code = MAD_FLOW_CONTINUE;
    } else {
        ret_code = MAD_FLOW_STOP;
    }

    return ret_code;

}

       我们设置的输入buff缓冲区的大小是8192字节，但是对于mp3文件来讲，不一定这8192个字节就刚好是若干个完整的帧，有可能会有若干字节是输入下一个帧的，所有要根据struct mad_stream中的两个指针，标示了缓冲区中的完整帧的起始地址：

unsigned char const *this_frame;    /* start
 of current frame */
unsigned char const *next_frame;    /* start
 of next frame */

那么

unproc_data_size = stream->bufend - stream->next_frame;

得到剩余的下一个帧的数据，并且需要将其从buff数组的尾部拷贝到头部，再从mp3文件中读取一部分字节拼凑成下一个8192字节，提交给库去解码，如此周而复始。

       4、此代码解码出来的pcm可以加上44字节的wav头文件，则可以用播放器正常播放。

 

     五、如何从网络socket获取相应数据，边解码边播放

     由于我的项目是要实现一个远程播放器的功能，即手机端的mp3源文件通过wifi传输到开发板上解码播放，所以，对于输入缓冲区的控制就不像操作文件那个，可以通过file结构体精确控制好读取的数据位置了，为此，做了些许修改。

     可以开两个线程，一个线程用于接收socket数据，一个用于解码播放。主要是缓冲区的控制，可以如此实现：将接收buff[]大小设置为8192*10字节，然后，解码input函数里面的buff[]的大小设置为8192*11字节，也就是说，多余了8192用来缓冲多余的下一帧字节的数据（因为mp3文件的帧不会超过8192字节），那么，区别于上面的思路，我们可以固定的让socket的buff[]接收8192*10字节的数据，如果解码的buff[]里面初次解码后有剩余的数据，仍然将其复制到解码buff[]的头部，只是这时候还是将socket的buff[]的8192*10字节的数据加到解码buff[]的刚刚拷贝的数据后面，所以，这里调用mad_stream_buffer(stream, buf, bsize)中的bsize就是8192*10+剩余的帧数据大小了。

 

相关参考：

1、作者：cqulpj  网址： http://cqulpj.blogbus.com/logs/68406670.html

                               http://cqulpj.blogbus.com/logs/68406669.html

2、作者：李素科  网址： http://www.ibm.com/developerworks/cn/linux/l-cn-libmadmp3player/index.html

3、作者：liky125 网址： http://blog.chinaunix.net/uid-26073752-id-2392553.html