ffmpeg-libopencore-amr 语音编码

AMR（Adaptive Multi-rate），自适应多速率语音编码器，主要用于移动设备的音频(GSM, 3G wcdma)，压缩比大，但相对其他的音频压缩格式音质差，多用于人声通话。

AMR又分为两种，一 种是AMR－NB（AMR-NarrowBind）窄频，语音带宽范围：300－3700Hz，8KHz采样频率；支持的输出bitrate有（4.75k,5.15k, 5.9k, 6.7k, 7.4k, 7.95k, 10.2k, 12.2k）, 肯定是bitrate越高音质越好了。

另外一种是AMR-WB（AMR WideBand）宽频，语音带宽范围50－7000Hz，16KHz采样频率。但考虑语音的短时相关性，每帧长度均为20ms。

 

AMR编码速率是可变的，每160个Samples编码一帧AMR数据，每帧中都要指明对应的bitrate,帧间bitrate可变。

分析一下ffmpeg中libopencore-amr.c中的amrnb的编码code，感觉它写的有问题，自己修改的思想，如code中的注释。

/*
 * AMR Audio decoder stub
 * Copyright (c) 2003 the ffmpeg project
 *
 * This file is part of FFmpeg.
 *
 * FFmpeg is free software; you can redistribute it and/or
 * modify it under the terms of the GNU Lesser General Public
 * License as published by the Free Software Foundation; either
 * version 2.1 of the License, or (at your option) any later version.
 *
 * FFmpeg is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * Lesser General Public License for more details.
 *
 * You should have received a copy of the GNU Lesser General Public
 * License along with FFmpeg; if not, write to the Free Software
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
 */

#include "libavutil/avstring.h"
#include "libavutil/channel_layout.h"
#include "libavutil/common.h"
#include "libavutil/opt.h"
#include "avcodec.h"
#include "audio_frame_queue.h"
#include "internal.h"

#include <opencore-amrnb/interf_dec.h>
#include <opencore-amrnb/interf_enc.h>

typedef struct AMRContext {
    AVClass *av_class;
    void *dec_state;
    void *enc_state;
    int   enc_bitrate;
    int   enc_mode;
    int   enc_dtx;
    int   enc_last_frame;
    AudioFrameQueue afq;
} AMRContext;

/* Common code for fixed and float version*/
typedef struct AMR_bitrates {
    int       rate;
    enum Mode mode;
} AMR_bitrates;

/* Match desired bitrate */
static int get_bitrate_mode(int bitrate, void *log_ctx)
{
    /* make the correspondance between bitrate and mode */
    static const AMR_bitrates rates[] = {
        { 4750, MR475 }, { 5150, MR515 }, {  5900, MR59  }, {  6700, MR67  },
        { 7400, MR74 },  { 7950, MR795 }, { 10200, MR102 }, { 12200, MR122 }
    };
    int i, best = -1, min_diff = 0;
    char log_buf[200];

    for (i = 0; i < 8; i++) {
        if (rates[i].rate == bitrate)
            return rates[i].mode;
        if (best < 0 || abs(rates[i].rate - bitrate) < min_diff) {
            best     = i;
            min_diff = abs(rates[i].rate - bitrate);
        }
    }
    /* no bitrate matching exactly, log a warning */
    snprintf(log_buf, sizeof(log_buf), "bitrate not supported: use one of ");
    for (i = 0; i < 8; i++)
        av_strlcatf(log_buf, sizeof(log_buf), "%.2fk, ", rates[i].rate    / 1000.f);
    av_strlcatf(log_buf, sizeof(log_buf), "using %.2fk", rates[best].rate / 1000.f);
    av_log(log_ctx, AV_LOG_WARNING, "%s\n", log_buf);

    return best;
}

static const AVOption options[] = {
    { "dtx", "Allow DTX (generate comfort noise)", offsetof(AMRContext, enc_dtx), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, 1, AV_OPT_FLAG_AUDIO_PARAM | AV_OPT_FLAG_ENCODING_PARAM },
    { NULL }
};

static const AVClass amrnb_class = {
    "libopencore_amrnb", av_default_item_name, options, LIBAVUTIL_VERSION_INT
};

static av_cold int amr_nb_encode_init(AVCodecContext *avctx)
{
    AMRContext *s = avctx->priv_data;

    if (avctx->sample_rate != 8000 && avctx->strict_std_compliance > FF_COMPLIANCE_UNOFFICIAL) {
        av_log(avctx, AV_LOG_ERROR, "Only 8000Hz sample rate supported\n");
        return AVERROR(ENOSYS);
    }

    if (avctx->channels != 1) {
        av_log(avctx, AV_LOG_ERROR, "Only mono supported\n");
        return AVERROR(ENOSYS);
    }

	/*8k samples rate, 每隔20ms 编码一次，所以编码的frame_size = 8000 * 20 /100 = 150*/
    avctx->frame_size  = 160;
    avctx->delay       =  50;
	/*加入一个队列，大概是保证每次丢给Encoder的都有160个Samples，但是感觉这个开源的写的不好*/
    ff_af_queue_init(avctx, &s->afq);

    s->enc_state = Encoder_Interface_init(s->enc_dtx);
    if (!s->enc_state) {
        av_log(avctx, AV_LOG_ERROR, "Encoder_Interface_init error\n");
        av_freep(&avctx->coded_frame);
        return -1;
    }

    s->enc_mode    = get_bitrate_mode(avctx->bit_rate, avctx);
    s->enc_bitrate = avctx->bit_rate;

    return 0;
}

static av_cold int amr_nb_encode_close(AVCodecContext *avctx)
{
    AMRContext *s = avctx->priv_data;

    Encoder_Interface_exit(s->enc_state);
    ff_af_queue_close(&s->afq);
    return 0;
}

static int amr_nb_encode_frame(AVCodecContext *avctx, AVPacket *avpkt,
                               const AVFrame *frame, int *got_packet_ptr)
{
    AMRContext *s = avctx->priv_data;
    int written, ret;
    int16_t *flush_buf = NULL;
    const int16_t *samples = frame ? (const int16_t *)frame->data[0] : NULL;

    if (s->enc_bitrate != avctx->bit_rate) {
        s->enc_mode    = get_bitrate_mode(avctx->bit_rate, avctx);
        s->enc_bitrate = avctx->bit_rate;
    }
	/*mode 7: max amr frame size is 32*/
	/*amr是比特率可以动态变化的，但是最大也只有的32个字节，所以分配32个字节*/
    if ((ret = ff_alloc_packet2(avctx, avpkt, 32)) < 0)
        return ret;

    if (frame) {
		/*amr encoder是要求每次输入160个samples,但是看这里的代码，好像小于160也没有关系，小于160了，copy到flushbuf中，后面直接相当于补0了。
		这个地方直接导致我输出的amr音质很差，因为我每次丢给它的samples个数远小于160
		据说ffmpeg中有audio fifo机制，但是不会用，索性自己写个buffer了，攒够160个samples再压缩，然后再换算一下输出的pts就好了*/
        if (frame->nb_samples < avctx->frame_size) {
            flush_buf = av_mallocz(avctx->frame_size * sizeof(*flush_buf));
            if (!flush_buf)
                return AVERROR(ENOMEM);
            memcpy(flush_buf, samples, frame->nb_samples * sizeof(*flush_buf));
            samples = flush_buf;
            if (frame->nb_samples < avctx->frame_size - avctx->delay)
                s->enc_last_frame = -1;
        }
		/*没有搞懂这个队列有什么用，不符合积攒够160个Sampels再压缩的原则*/
        if ((ret = ff_af_queue_add(&s->afq, frame)) < 0) {

			av_log(avctx, AV_LOG_ERROR, "too small samples need to return\n");
            av_freep(&flush_buf);
            return ret;
        }

		av_log(avctx, AV_LOG_INFO, "going to encode frame \n");
    } else {
		av_log(avctx, AV_LOG_INFO, "didnot get frame\n");
        if (s->enc_last_frame < 0)
            return 0;
        flush_buf = av_mallocz(avctx->frame_size * sizeof(*flush_buf));
        if (!flush_buf)
            return AVERROR(ENOMEM);
        samples = flush_buf;
        s->enc_last_frame = -1;
    }

    written = Encoder_Interface_Encode(s->enc_state, s->enc_mode, samples,
                                       avpkt->data, 0);
    av_dlog(avctx, "amr_nb_encode_frame encoded %u bytes, bitrate %u, first byte was %#02x\n",
            written, s->enc_mode, avpkt->data[0]);

    /* Get the next frame pts/duration */
    ff_af_queue_remove(&s->afq, avctx->frame_size, &avpkt->pts,
                       &avpkt->duration);
	av_log(avctx, AV_LOG_INFO,
		"frame_size %d pts :%lld, duration :%d",
		avctx->frame_size, avpkt->pts, avpkt->duration);

    avpkt->size = written;
    *got_packet_ptr = 1;
    av_freep(&flush_buf);

    return 0;
}

AVCodec ff_libopencore_amrnb_encoder = {
    .name           = "libopencore_amrnb",
    .long_name      = NULL_IF_CONFIG_SMALL("OpenCORE AMR-NB (Adaptive Multi-Rate Narrow-Band)"),
    .type           = AVMEDIA_TYPE_AUDIO,
    .id             = AV_CODEC_ID_AMR_NB,
    .priv_data_size = sizeof(AMRContext),
    .init           = amr_nb_encode_init,
    .encode2        = amr_nb_encode_frame,
    .close          = amr_nb_encode_close,
    .capabilities   = CODEC_CAP_DELAY | CODEC_CAP_SMALL_LAST_FRAME,
    .sample_fmts    = (const enum AVSampleFormat[]){ AV_SAMPLE_FMT_S16,
                                                     AV_SAMPLE_FMT_NONE },
    .priv_class     = &amrnb_class,
};

后来还是要自己修改下，才能适应自己的工程。修改思想见code中的注释。