Android平台RTSP转RTMP推送之采集麦克风音频转发

技术背景

RTSP转RTMP推送，好多开发者第一想到的是采用ffmpeg命令行的形式，如果对ffmpeg比较熟，而且产品不要额外的定制和更高阶的要求，未尝不可，如果对产品稳定性、时延、断网重连等有更高的技术诉求，比较好的办法，还是采用我们的技术实现。

技术实现

以大牛直播SDK的多路RTSP转RTMP推送模块为例，首先拉取RTSP流，把未解码的H.264/H.265、AAC/PCMA/PCMU数据回调上来，然后通过调用推送模块的编码后数据接口，同步转发出去，整体下来，几无多少延迟。如果需要把数据投递到轻量级RTSP服务也可以。系统设计架构图如下：

1. 拉流：通过RTSP直播播放SDK的数据回调接口，拿到音视频数据；

2. 转推：通过RTMP直播推送SDK的编码后数据输入接口，把回调上来的数据，传给RTMP直播推送模块，实现RTSP数据流到RTMP服务器的转发；

3. 录像：如果需要录像，借助RTSP直播播放SDK，拉到音视频数据后，直接存储MP4文件即可；

4. 快照：如果需要实时快照，拉流后，解码调用播放端快照接口，生成快照，因为快照涉及到video数据解码，如无必要，可不必开启，不然会额外消耗性能。

5. 拉流预览：如需预览拉流数据，只要调用播放端的播放接口，即可实现拉流数据预览；

6. 数据转AAC后转发：考虑到好多监控设备出来的音频可能是PCMA/PCMU的，如需要更通用的音频格式，可以转AAC后，在通过RTMP推送；

7. 转推RTMP实时静音：只需要在传audio数据的地方，加个判断即可；

8. 拉流速度反馈：通过RTSP播放端的实时码率反馈event，拿到实时带宽占用即可；

9. 整体网络状态反馈：考虑到有些摄像头可能会临时或异常关闭，RTMP服务器亦是，可以通过推拉流的event回调状态，查看那整体网络情况，如此界定：是拉不到流，还是推不到RTMP服务器。

多路RTMP/RTSP转RTMP推送模块功能支持：

1. 支持拉取rtmp流；
2. 支持拉取rtsp流；
3. Windows支持本地flv文件转发(支持制定文件位置转发，或转发过程中seek)；
4. 支持本地预览；
5. 支持转发过程中，实时静音；
6. 支持转发过程中，切换rtmp/rtsp url，此外，windows平台还支持切换本地flv文件；
7. 支持录像模块扩展，可边转发边录制，每个文件录制开始结束，均有状态回馈；
8. 支持内网RTSP网关模块扩展，拉取的流数据，可以流入到内网RTSP网关模块，对外微型RTSP媒体流服务（RTSP url），便于内网访问；
9. 音频：AAC，并支持拉流后的音频(PCMU/PCMA,Speex等)转AAC后再转发；
10. 视频：H.264、H.265，支持h265转发(rtsp/rtmp h265转rtmp h265推送)；

上述实现，2016年我们已经非常成熟，本次要谈的，是开发者实际场景用到的一个技术需求，如何实现视频用RTSP数据源获取到的，音频采集麦克风的数据。

废话不多说，上代码：

先说开始拉流、停止拉流设计如下，如果是用rtsp的audio，那么我们就开启audio数据的回调，如果采用麦克风的，这里只要开video的即可。

/** SmartRelayDemo.java* Created by daniusdk.com* weChat: xinsheng120*/
private boolean StartPull()
{if ( isPulling )return false;if(!isPlaying){if (!OpenPullHandle())return false;}if(audio_opt_ == 2){libPlayer.SmartPlayerSetAudioDataCallback(player_handle_, new PlayerAudioDataCallback(stream_publisher_));}if(video_opt_ == 2){libPlayer.SmartPlayerSetVideoDataCallback(player_handle_, new PlayerVideoDataCallback(stream_publisher_));}int is_pull_trans_code  = 1;libPlayer.SmartPlayerSetPullStreamAudioTranscodeAAC(player_handle_, is_pull_trans_code);int startRet = libPlayer.SmartPlayerStartPullStream(player_handle_);if (startRet != 0) {Log.e(TAG, "Failed to start pull stream!");if(!isPlaying){releasePlayerHandle();}return false;}isPulling = true;return true;
}private void StopPull()
{if ( !isPulling )return;isPulling = false;if (null == libPlayer || 0 == player_handle_)return;libPlayer.SmartPlayerStopPullStream(player_handle_);if ( !isPlaying){releasePlayerHandle();}
}

OpenPullHandle()实现逻辑如下，常规的参数设置，和event callback设置等。

private boolean OpenPullHandle()
{//playbackUrl可自定义playbackUrl = "rtsp://admin:daniulive12345@192.168.0.120:554/h264/ch1/main/av_stream";if (playbackUrl == null) {Log.e(TAG, "playback URL is null...");return false;}player_handle_ = libPlayer.SmartPlayerOpen(context_);if (player_handle_ == 0) {Log.e(TAG, "playerHandle is null..");return false;}libPlayer.SetSmartPlayerEventCallbackV2(player_handle_,new EventHandlePlayerV2());libPlayer.SmartPlayerSetBuffer(player_handle_, playBuffer);// set report download speedlibPlayer.SmartPlayerSetReportDownloadSpeed(player_handle_, 1, 2);//设置RTSP超时时间int rtsp_timeout = 10;libPlayer.SmartPlayerSetRTSPTimeout(player_handle_, rtsp_timeout);//设置RTSP TCP/UDP模式自动切换int is_auto_switch_tcp_udp = 1;libPlayer.SmartPlayerSetRTSPAutoSwitchTcpUdp(player_handle_, is_auto_switch_tcp_udp);libPlayer.SmartPlayerSaveImageFlag(player_handle_, 1);// It only used when playback RTSP stream..//libPlayer.SmartPlayerSetRTSPTcpMode(playerHandle, 1);libPlayer.SmartPlayerSetUrl(player_handle_, playbackUrl);return true;
}

拉流后，转推RTMP的设计如下：

btnRTMPPusher.setOnClickListener(new Button.OnClickListener() {// @Overridepublic void onClick(View v) {if (stream_publisher_.is_rtmp_publishing()) {stopPush();btnRTMPPusher.setText("推送RTMP");return;}Log.i(TAG, "onClick start push rtmp..");InitAndSetConfig();String rtmp_pusher_url = "rtmp://192.168.0.104:1935/hls/stream1";//String rtmp_pusher_url = relayStreamUrl;if (!stream_publisher_.SetURL(rtmp_pusher_url))Log.e(TAG, "Failed to set publish stream URL..");boolean start_ret = stream_publisher_.StartPublisher();if (!start_ret) {stream_publisher_.try_release();Log.e(TAG, "Failed to start push stream..");return;}startAudioRecorder();btnRTMPPusher.setText("停止推送");}
});

InitAndSetConfig()设计如下：

private void InitAndSetConfig() {if (null == libPublisher)return;if (!stream_publisher_.empty())return;Log.i(TAG, "InitAndSetConfig video width: " + video_width_ + ", height" + video_height_);long handle = libPublisher.SmartPublisherOpen(context_, audio_opt_, video_opt_,  video_width_, video_height_);if (0==handle) {Log.e(TAG, "sdk open failed!");return;}Log.i(TAG, "publisherHandle=" + handle);int fps = 25;int gop = fps * 3;initialize_publisher(libPublisher, handle, video_width_, video_height_, fps, gop);stream_publisher_.set(libPublisher, handle);
}

这里可以看到，我们在转推RTMP的时候，调用了startAudioRecorder()来做麦克风的采集：

void startAudioRecorder() {if(audio_opt_ != 1)return;if (audio_recorder_ != null)return;audio_recorder_ = new NTAudioRecordV2(this);Log.i(TAG, "startAudioRecorder call audio_recorder_.start()+++...");audio_recorder_callback_ = new NTAudioRecordV2CallbackImpl(stream_publisher_, null);audio_recorder_.AddCallback(audio_recorder_callback_);if (!audio_recorder_.Start(is_pcma_ ? 8000 : 44100, 1) ) {audio_recorder_.RemoveCallback(audio_recorder_callback_);audio_recorder_callback_ = null;audio_recorder_ = null;Log.e(TAG, "startAudioRecorder start failed.");}else {Log.i(TAG, "startAudioRecorder call audio_recorder_.start() OK---...");}
}void stopAudioRecorder() {if (null == audio_recorder_)return;Log.i(TAG, "stopAudioRecorder+++");audio_recorder_.Stop();if (audio_recorder_callback_ != null) {audio_recorder_.RemoveCallback(audio_recorder_callback_);audio_recorder_callback_ = null;}audio_recorder_ = null;Log.i(TAG, "stopAudioRecorder---");
}

采集到的audio回调上来后，我们调RTMP推送接口，把数据投递下去即可：

private static class NTAudioRecordV2CallbackImpl implements NTAudioRecordV2Callback {private WeakReference<LibPublisherWrapper> publisher_0_;private WeakReference<LibPublisherWrapper> publisher_1_;public NTAudioRecordV2CallbackImpl(LibPublisherWrapper publisher_0, LibPublisherWrapper publisher_1) {if (publisher_0 != null)publisher_0_ = new WeakReference<>(publisher_0);if (publisher_1 != null)publisher_1_ = new WeakReference<>(publisher_1);}private final LibPublisherWrapper get_publisher_0() {if (publisher_0_ !=null)return publisher_0_.get();return null;}private final LibPublisherWrapper get_publisher_1() {if (publisher_1_ != null)return publisher_1_.get();return null;}@Overridepublic void onNTAudioRecordV2Frame(ByteBuffer data, int size, int sampleRate, int channel, int per_channel_sample_number) {//Log.i(TAG, "onNTAudioRecordV2Frame size=" + size + " sampleRate=" + sampleRate + " channel=" + channel//			 + " per_channel_sample_number=" + per_channel_sample_number);LibPublisherWrapper publisher_0 = get_publisher_0();if (publisher_0 != null)publisher_0.OnPCMData(data, size, sampleRate, channel, per_channel_sample_number);LibPublisherWrapper publisher_1 = get_publisher_1();if (publisher_1 != null)publisher_1.OnPCMData(data, size, sampleRate, channel, per_channel_sample_number);}
}

编码后的视频投递设计如下：

class PlayerVideoDataCallback implements NTVideoDataCallback
{private WeakReference<LibPublisherWrapper> publisher_;private int video_buffer_size = 0;private ByteBuffer video_buffer_ = null;public PlayerVideoDataCallback(LibPublisherWrapper publisher) {if (publisher != null)publisher_ = new WeakReference<>(publisher);}@Overridepublic ByteBuffer getVideoByteBuffer(int size){if( size < 1 ){return null;}if ( size <= video_buffer_size &&  video_buffer_ != null ){return  video_buffer_;}video_buffer_size = size + 1024;video_buffer_size = (video_buffer_size+0xf) & (~0xf);video_buffer_ = ByteBuffer.allocateDirect(video_buffer_size);return video_buffer_;}public void onVideoDataCallback(int ret, int video_codec_id, int sample_size, int is_key_frame, long timestamp, int width, int height, long presentation_timestamp){if ( video_buffer_ == null)return;LibPublisherWrapper publisher = publisher_.get();if (null == publisher)return;if (!publisher.is_publishing())return;video_buffer_.rewind();publisher.PostVideoEncodedData(video_codec_id, video_buffer_, sample_size, is_key_frame, timestamp, presentation_timestamp);}
}

总结

从我发的Android平台RTSP转RTMP推送的demo界面，可以看到，这个demo，不是单纯的RTSP转RTMP推送的，还可以实现RTSP流获取后，回调上来解码后的数据，然后添加动态水印或其他处理后，把video数据二次编码推送出去。或者audio数据二次处理。

此外，还可以实现拉流的数据预览播放、把数据注入到轻量级RTSP服务模块，然后二次编码的数据，本地录像、快照等。一个好的RTSP转RTMP推送的模块，一定要足够的灵活，扩展性好，才能很快的实现客户的技术诉求。以上抛砖引玉，感兴趣的开发者，可以跟我单独探讨。