正在制作某物品,现在做到音频部分了。
原本要采用 SDL2_mixer 的,不过实验结果表明其失真非常严重,还带有大量的电噪声。不知道是不是我打开的方式不对……
一气之下去看 OpenAL,结果吃了闭门羹(维护中,只有 mailing list 和 specification)。转投 FMOD,不过又考虑到其授权方式,还是放弃了。最终回到 OpenAL。使用的是 OpenAL-Soft。
OpenAL 呢,好的方面是开源+授权,坏的方面……呃,至少在刚刚的测试中,代码维护甚至没有 SDL 好。直接编译 .c 示例失败,耍小聪明改成 .cpp 拿去编译才成功。
在接下来的代码中,需要用到 OpenAL-Soft(1.15.1)和 FFmpeg。
看 OpenAL-Soft 自带的示例 alstream.c。为了方便起见,接下来的 C 源代码文件全部改成 C++ 源代码文件去……同时不要忘了在 FFmpeg 的头文件上下加 extern “C”!(为什么他们不考虑这一点?)
好,编译示例,运行。(注意,各种 dependencies 这里就不提了。)随便选择一个含有音频的、可以被 FFmpeg 解码的文件。
不对啊!很有可能出现以下错误信息:
Opened “OpenAL Soft”
AL_SOFT_buffer_samples supported!
Unsupported ffmpeg sample format: s16p
Error getting audio info for 01.mpg
Done.
这是……怎么回事?经过测试,SDL_mixer
可以播放同一个文件,不过正如之前所说的,失真&噪声。看其采样格式:S16P(Signed 16-bit, Planar←平面?)。再看源代码,S16(Signed 16-bit)是支持的。(当然,如果强制将那几个 if 修改一下的话,你会听到神奇的东西……)S16 和 S16P 的不同点是在于数据的排列方式,前者是相邻连续排列,后者是分离排列。但是现在有相当多的音频文件采用 planar 的方案,不仅是 S16,U8、S32、F32、F64 都有对应的 planar 方式。现在,目标就是:让这个示例支持 planar。
思路很简单。我的上一篇随笔中,有一个 AudioResampling()
函数,这里直接拿来用吧!(秉持拿来主义!鲁迅先生不谢。)
接下来就是好戏了。
又试验了一下,播放 U8/Mono 的时候出现崩溃,不知道原因。调试的时候内存是越界的。
先是添加对 libswresample 和 libavutil(要用到 opt_*
函数)的包含(别忘了添加对应的库):
#ifdef __cplusplus
extern "C" {
#endif
#include "libavutil/opt.h"
#include "libswresample/swresample.h"
#ifdef __cplusplus
}
#endif
然后是修改 MyStream
的定义:
struct MyStream {
AVCodecContext *CodecCtx;
int StreamIdx;
struct PacketList *Packets;
AVFrame *Frame;
// FrameData 没什么用了,不过为了保持代码结构,还是保留下来,其作用由 FrameBuffer 代替
const uint8_t *FrameData;
const uint8_t FrameBuffer[FRAME_BUFFER_SIZE];
size_t FrameDataSize;
FilePtr parent;
};
可以先定义一下 FRAME_BUFFER_SIZE
:
// MP3 每一帧的大小是4608,所以如果设定成4096(一般音频可以播放)的话会造成溢出、崩溃
#define FRAME_BUFFER_SIZE (4800)
直接插入 AudioResampling()
函数(如果对这错误的时态感到别扭,改一下就好了),添加重采样支持:
static int AudioResampling(AVCodecContext * audio_dec_ctx,
AVFrame * pAudioDecodeFrame,
int out_sample_fmt,
int out_channels,
int out_sample_rate,
uint8_t* out_buf)
{
SwrContext * swr_ctx = NULL;
int data_size = 0;
int ret = 0;
int64_t src_ch_layout = audio_dec_ctx->channel_layout;
int64_t dst_ch_layout = AV_CH_LAYOUT_STEREO;
int dst_nb_channels = 0;
int dst_linesize = 0;
int src_nb_samples = 0;
int dst_nb_samples = 0;
int max_dst_nb_samples = 0;
uint8_t **dst_data = NULL;
int resampled_data_size = 0;
swr_ctx = swr_alloc();
if (!swr_ctx)
{
printf("swr_alloc error \n");
return -1;
}
src_ch_layout = (audio_dec_ctx->channels ==
av_get_channel_layout_nb_channels(audio_dec_ctx->channel_layout)) ?
audio_dec_ctx->channel_layout :
av_get_default_channel_layout(audio_dec_ctx->channels);
if (out_channels == 1)
{
dst_ch_layout = AV_CH_LAYOUT_MONO;
//printf("dst_ch_layout: AV_CH_LAYOUT_MONO\n");
}
else if (out_channels == 2)
{
dst_ch_layout = AV_CH_LAYOUT_STEREO;
//printf("dst_ch_layout: AV_CH_LAYOUT_STEREO\n");
}
else
{
dst_ch_layout = AV_CH_LAYOUT_SURROUND;
//printf("dst_ch_layout: AV_CH_LAYOUT_SURROUND\n");
}
if (src_ch_layout <= 0)
{
printf("src_ch_layout error \n");
return -1;
}
src_nb_samples = pAudioDecodeFrame->nb_samples;
if (src_nb_samples <= 0)
{
printf("src_nb_samples error \n");
return -1;
}
av_opt_set_int(swr_ctx, "in_channel_layout", src_ch_layout, 0);
av_opt_set_int(swr_ctx, "in_sample_rate", audio_dec_ctx->sample_rate, 0);
av_opt_set_sample_fmt(swr_ctx, "in_sample_fmt", audio_dec_ctx->sample_fmt, 0);
av_opt_set_int(swr_ctx, "out_channel_layout", dst_ch_layout, 0);
av_opt_set_int(swr_ctx, "out_sample_rate", out_sample_rate, 0);
av_opt_set_sample_fmt(swr_ctx, "out_sample_fmt", (AVSampleFormat)out_sample_fmt, 0);
if ((ret = swr_init(swr_ctx)) < 0) {
printf("Failed to initialize the resampling context\n");
return -1;
}
max_dst_nb_samples = dst_nb_samples = av_rescale_rnd(src_nb_samples,
out_sample_rate, audio_dec_ctx->sample_rate, AV_ROUND_UP);
if (max_dst_nb_samples <= 0)
{
printf("av_rescale_rnd error \n");
return -1;
}
dst_nb_channels = av_get_channel_layout_nb_channels(dst_ch_layout);
ret = av_samples_alloc_array_and_samples(&dst_data, &dst_linesize, dst_nb_channels,
dst_nb_samples, (AVSampleFormat)out_sample_fmt, 0);
if (ret < 0)
{
printf("av_samples_alloc_array_and_samples error \n");
return -1;
}
dst_nb_samples = av_rescale_rnd(swr_get_delay(swr_ctx, audio_dec_ctx->sample_rate) +
src_nb_samples, out_sample_rate, audio_dec_ctx->sample_rate, AV_ROUND_UP);
if (dst_nb_samples <= 0)
{
printf("av_rescale_rnd error \n");
return -1;
}
if (dst_nb_samples > max_dst_nb_samples)
{
av_free(dst_data[0]);
ret = av_samples_alloc(dst_data, &dst_linesize, dst_nb_channels,
dst_nb_samples, (AVSampleFormat)out_sample_fmt, 1);
max_dst_nb_samples = dst_nb_samples;
}
if (swr_ctx)
{
ret = swr_convert(swr_ctx, dst_data, dst_nb_samples,
(const uint8_t **)pAudioDecodeFrame->data, pAudioDecodeFrame->nb_samples);
if (ret < 0)
{
printf("swr_convert error \n");
return -1;
}
resampled_data_size = av_samples_get_buffer_size(&dst_linesize, dst_nb_channels,
ret, (AVSampleFormat)out_sample_fmt, 1);
if (resampled_data_size < 0)
{
printf("av_samples_get_buffer_size error \n");
return -1;
}
}
else
{
printf("swr_ctx null error \n");
return -1;
}
memcpy(out_buf, dst_data[0], resampled_data_size);
if (dst_data)
{
av_freep(&dst_data[0]);
}
av_freep(&dst_data);
dst_data = NULL;
if (swr_ctx)
{
swr_free(&swr_ctx);
}
return resampled_data_size;
}
修改 getAVAudioData()
函数:
uint8_t *getAVAudioData(StreamPtr stream, size_t *length)
{
int got_frame;
int len;
if(length) *length = 0;
if(!stream || stream->CodecCtx->codec_type != AVMEDIA_TYPE_AUDIO)
return NULL;
next_packet:
if(!stream->Packets && !getNextPacket(stream->parent, stream->StreamIdx))
return NULL;
/* Decode some data, and check for errors */
avcodec_get_frame_defaults(stream->Frame);
while((len=avcodec_decode_audio4(stream->CodecCtx, stream->Frame,
&got_frame, &stream->Packets->pkt)) < 0)
{
struct PacketList *self;
/* Error? Drop it and try the next, I guess... */
self = stream->Packets;
stream->Packets = self->next;
av_free_packet(&self->pkt);
av_free(self);
if(!stream->Packets)
goto next_packet;
}
if(len < stream->Packets->pkt.size)
{
/* Move the unread data to the front and clear the end bits */
int remaining = stream->Packets->pkt.size - len;
memmove(stream->Packets->pkt.data, &stream->Packets->pkt.data[len],
remaining);
memset(&stream->Packets->pkt.data[remaining], 0,
stream->Packets->pkt.size - remaining);
stream->Packets->pkt.size -= len;
}
else
{
struct PacketList *self;
self = stream->Packets;
stream->Packets = self->next;
av_free_packet(&self->pkt);
av_free(self);
}
if(!got_frame || stream->Frame->nb_samples == 0)
goto next_packet;
// 在这里插入重新采样代码
*length = AudioResampling(stream->CodecCtx, stream->Frame, AV_SAMPLE_FMT_S16, stream->Frame->channels, stream->Frame->sample_rate, const_cast<uint8_t *>(stream->FrameBuffer));
/* Set the output buffer size */
/*
*length = av_samples_get_buffer_size(NULL, stream->CodecCtx->channels,
stream->Frame->nb_samples,
stream->CodecCtx->sample_fmt, 1);
return stream->Frame->data[0];
*/
return const_cast<uint8_t *>(stream->FrameBuffer);
}
最后是 getAVAudioInfo()
函数,我们要让它允许 planar 音频输入:
int getAVAudioInfo(StreamPtr stream, ALuint *rate, ALenum *channels, ALenum *type)
{
if(!stream || stream->CodecCtx->codec_type != AVMEDIA_TYPE_AUDIO)
return 1;
/* Get the sample type for OpenAL given the format detected by ffmpeg. */
if(stream->CodecCtx->sample_fmt == AV_SAMPLE_FMT_U8 || stream->CodecCtx->sample_fmt == AV_SAMPLE_FMT_U8P)
*type = AL_UNSIGNED_BYTE_SOFT;
else if (stream->CodecCtx->sample_fmt == AV_SAMPLE_FMT_S16 || stream->CodecCtx->sample_fmt == AV_SAMPLE_FMT_S16P)
*type = AL_SHORT_SOFT;
else if(stream->CodecCtx->sample_fmt == AV_SAMPLE_FMT_S32 || stream->CodecCtx->sample_fmt == AV_SAMPLE_FMT_S32P)
*type = AL_INT_SOFT;
else if(stream->CodecCtx->sample_fmt == AV_SAMPLE_FMT_FLT || stream->CodecCtx->sample_fmt == AV_SAMPLE_FMT_FLTP)
*type = AL_FLOAT_SOFT;
else if(stream->CodecCtx->sample_fmt == AV_SAMPLE_FMT_DBL || stream->CodecCtx->sample_fmt == AV_SAMPLE_FMT_DBLP)
*type = AL_DOUBLE_SOFT;
else
{
fprintf(stderr, "Unsupported ffmpeg sample format: %s\n",
av_get_sample_fmt_name(stream->CodecCtx->sample_fmt));
return 1;
}
/* Get the OpenAL channel configuration using the channel layout detected
* by ffmpeg. NOTE: some file types may not specify a channel layout. In
* that case, one must be guessed based on the channel count. */
if(stream->CodecCtx->channel_layout == AV_CH_LAYOUT_MONO)
*channels = AL_MONO_SOFT;
else if(stream->CodecCtx->channel_layout == AV_CH_LAYOUT_STEREO)
*channels = AL_STEREO_SOFT;
else if(stream->CodecCtx->channel_layout == AV_CH_LAYOUT_QUAD)
*channels = AL_QUAD_SOFT;
else if(stream->CodecCtx->channel_layout == AV_CH_LAYOUT_5POINT1_BACK)
*channels = AL_5POINT1_SOFT;
else if(stream->CodecCtx->channel_layout == AV_CH_LAYOUT_7POINT1)
*channels = AL_7POINT1_SOFT;
else if(stream->CodecCtx->channel_layout == 0)
{
/* Unknown channel layout. Try to guess. */
if(stream->CodecCtx->channels == 1)
*channels = AL_MONO_SOFT;
else if(stream->CodecCtx->channels == 2)
*channels = AL_STEREO_SOFT;
else
{
fprintf(stderr, "Unsupported ffmpeg raw channel count: %d\n",
stream->CodecCtx->channels);
return 1;
}
}
else
{
char str[1024];
av_get_channel_layout_string(str, sizeof(str), stream->CodecCtx->channels,
stream->CodecCtx->channel_layout);
fprintf(stderr, "Unsupported ffmpeg channel layout: %s\n", str);
return 1;
}
*rate = stream->CodecCtx->sample_rate;
return 0;
}
嗯,基本上就可以了。现在播放的话,对应的 planar 是不会显示出来的,因为显示调用的是 alhelpers.cpp
的 GetFormat()
,而它是按照 OpenAL 的格式输出的。
不过这不影响播放嘛。