目前在做的在线直播教室,需要将老师分享的屏幕和老师的声音、学生的声音录制为一个视频文件,以便学生上课后还可以再看回放。

直播服务我们采用的是腾讯的视频服务,有现成的 SDK 可以用。但 SDK 自带的录制接口满足不了我们的需求,考察了 ffmpeg 和 GStreamer 后,决定在项目中使用 GStreamer 来实现。

在开始编写代码以前,先用命令行进行测试,命令行如下:

gst-launch-1.0.exe -v --gst-debug-level=4 flvmux name=mux ! tee name=t ! queue ! filesink name=file location=test.flv 
t. ! queue ! rtmpsink location="rtmp://live.abc.com/live/........" 
adder name=mix ! queue ! audiorate ! audioconvert ! voaacenc ! mux.audio 
videotestsrc name=screen_src ! queue ! videorate ! x264enc ! mux.video 
audiotestsrc name=send_audio_src ! queue ! audiorate ! mix. 
audiotestsrc wave=5 name=receive_audio_src ! queue ! audiorate ! mix.

命令看起来有点复杂,但其实逻辑挺简单的,看下面这个逻辑图就比较容易理解了(点击查看大图)


android 视频混流播放其中一个声音_操作系统


上面的命令中的推流地址需要替换为你的推流地址,如果没有,可以先把下面这部分内容去掉不推流




t. ! queue ! rtmpsink location="rtmp://live.abc.com/live/........"

另外这个命令是在 Windows 中的 MINGW64 的 bash 环境里面运行的, 如果在 Windows 的 cmd 环境中运行,把每行最后的  和换行去掉就可以了。

上面命令中用到了几个关键的插件,分别解释一下:

adder: 音频混流,将两路音频混为一路

voaacenc:音频编码,将原始的音频流编码为 aac 格式

x264enc:视频编码,将原始的视频流编码为 h264 格式

flvmux:flv 组装,将视频和音频组装在一起

tee:分流器,将一路输入变为两路输出,以分别进行后续的处理。一路保存为文件,一路进行推流

filesink:文件存储,将输入数据存储到指定的文件中

rtmpsink:推流,将输入数据推流到指定的视频服务器

接下来进行代码实现,我们的项目是 QT C++ 项目,目前仅在 Windows 平台使用

将 gstream 管道的初始化等放在 GStreamProcess 类中,代码如下:

GStreamProcess.h

#pragma once

#include "stdafx.h"

#include <gst/gst.h>
#include <gst/app/gstappsrc.h>
#include <gst/base/gstbaseparse.h>

typedef struct _AppSrcOption AppSrcOption;

struct _AppSrcOption
{
    _AppSrcOption()
        : pipeline(nullptr)
        , shareScreenAppsrc(nullptr)
        , sendAudioAppsrc(nullptr)
        , bus(nullptr)
        , gloop(nullptr)
        , width(0)
        , height(0)
    {}

    GstElement *pipeline;
    GstElement *shareScreenAppsrc;
    GstElement *sendAudioAppsrc;
    GstElement *receiveAudioAppsrc;
    GstElement *rtmp;
    GstBus *bus;
    GMainLoop *gloop;

    QString recordFileName;

    iLiveSucCallback sucCallback;
    iLiveErrCallback errCallback;
    void* callbackData;

    uint width;
    uint height;

    QString pushStreamUrl;
};

//int gstreamerInit(AppSrcOption *app, int argc, char *argv[]);


class GStreamProcess : public QThread
{
    Q_OBJECT

public:
    AppSrcOption* app;

protected:      
    void run() Q_DECL_OVERRIDE;

signals:
    void resultReady(const QString &s);
};

GStreamProcess.cpp

#include "stdafx.h"

GST_DEBUG_CATEGORY(appsrc_pipeline_debug);
#define GST_CAT_DEFAULT appsrc_pipeline_debug

static gboolean
bus_message(GstBus * bus, GstMessage * message, AppSrcOption * app)
{
    GST_DEBUG("got message %s",
        gst_message_type_get_name(GST_MESSAGE_TYPE(message)));

    switch (GST_MESSAGE_TYPE(message)) {
    case GST_MESSAGE_ERROR: {
        GError *err = NULL;
        gchar *dbg_info = NULL;

        gst_message_parse_error(message, &err, &dbg_info);

        gchar* elename = GST_OBJECT_NAME(message->src);

        g_printerr("ERROR from element %s: %s\n",
            elename, err->message);
        g_printerr("Debugging info: %s\n", (dbg_info) ? dbg_info : "none");

        app->errCallback(-90001, err->message, app->callbackData);

        g_error_free(err);
        g_free(dbg_info);
        g_main_loop_quit(app->gloop);
        break;
    }
    case GST_MESSAGE_EOS: {
        g_main_loop_quit(app->gloop);
        break;
    }
    default:
        break;
    }

    return TRUE;
}

void GStreamProcess::run()
{
    GError *error = NULL;

    int argc = 1;

    char *mock[1] = {"empty"};
    char **argv[1];
    *argv = mock;

    gst_init(&argc, argv);

    GST_DEBUG_CATEGORY_INIT(appsrc_pipeline_debug, "appsrc-pipeline", 0,
        "appsrc pipeline example");

    app->gloop = g_main_loop_new(NULL, TRUE);

    GstElement *pipeline = gst_parse_launch("flvmux name=mux ! queue ! tee name=t ! queue ! filesink name=file t. ! queue ! rtmpsink name=rtmp adder name=mix ! queue ! audiorate ! audioconvert ! voaacenc ! mux.audio appsrc name=screen_src ! queue ! videorate ! x264enc ! mux.video appsrc name=send_audio_src ! queue ! audiorate ! mix. appsrc wave=5 name=receive_audio_src ! queue ! audiorate ! mix.", NULL);
    g_assert(pipeline);

    app->bus = gst_pipeline_get_bus(GST_PIPELINE(pipeline));
    g_assert(app->bus);

    /* add watch for messages */
    gst_bus_add_watch(app->bus, (GstBusFunc)bus_message, app);


    /* 设置 screen src 属性 */

    app->shareScreenAppsrc = gst_bin_get_by_name(GST_BIN(pipeline), "screen_src");
    g_assert(app->shareScreenAppsrc);

    GstCaps *caps = gst_caps_new_simple("video/x-raw",
        "format", G_TYPE_STRING, "I420",
        "width", G_TYPE_INT, app->width,
        "height", G_TYPE_INT, app->height,
        "framerate", GST_TYPE_FRACTION, 15, 1,
        NULL);

    gst_app_src_set_caps(GST_APP_SRC(app->shareScreenAppsrc), caps);

    g_object_set(app->shareScreenAppsrc, "format", GST_FORMAT_TIME, NULL);
    g_object_set(app->shareScreenAppsrc, "is-live", TRUE, NULL);


    /* 设置 send audio src 属性 */

    app->sendAudioAppsrc = gst_bin_get_by_name(GST_BIN(pipeline), "send_audio_src");
    g_assert(app->sendAudioAppsrc);

    caps = gst_caps_new_simple("audio/x-raw",
        "format", G_TYPE_STRING, "S16LE",
        "layout", G_TYPE_STRING, "interleaved",
        "channels", G_TYPE_INT, 2,
        "rate", G_TYPE_INT, 48000,
        NULL);

    gst_app_src_set_caps(GST_APP_SRC(app->sendAudioAppsrc), caps);
    g_object_set(app->sendAudioAppsrc, "format", GST_FORMAT_TIME, NULL);
    g_object_set(app->sendAudioAppsrc, "is-live", TRUE, NULL);


    /* 设置 receive audio src 属性 */

    app->receiveAudioAppsrc = gst_bin_get_by_name(GST_BIN(pipeline), "receive_audio_src");
    g_assert(app->receiveAudioAppsrc);

    caps = gst_caps_new_simple("audio/x-raw",
        "format", G_TYPE_STRING, "S16LE",
        "layout", G_TYPE_STRING, "interleaved",
        "channels", G_TYPE_INT, 2,
        "rate", G_TYPE_INT, 48000,
        NULL);

    gst_app_src_set_caps(GST_APP_SRC(app->receiveAudioAppsrc), caps);
    g_object_set(app->receiveAudioAppsrc, "format", GST_FORMAT_TIME, NULL);
    g_object_set(app->receiveAudioAppsrc, "is-live", TRUE, NULL);


    /* 设置 filesink 属性 */

    GstElement *filesink = gst_bin_get_by_name(GST_BIN(pipeline), "file");
    g_assert(filesink);

    g_object_set(G_OBJECT(filesink), "location", app->recordFileName.toStdString().c_str(), NULL);


    /* 设置 rtmp 属性 */

    GstElement *rtmp = gst_bin_get_by_name(GST_BIN(pipeline), "rtmp");
    g_assert(rtmp);

    g_object_set(G_OBJECT(rtmp), "location", app->pushStreamUrl.toStdString().c_str(), NULL);

    /* go to playing */
    gst_element_set_state(pipeline, GST_STATE_PLAYING);

    //GST_DEBUG_BIN_TO_DOT_FILE_WITH_TS(GST_BIN(pipeline), GST_DEBUG_GRAPH_SHOW_ALL, "pipeline_dot");

    app->pipeline = pipeline;

    app->sucCallback(app->callbackData);

    g_main_loop_run(app->gloop);

    GST_DEBUG("stopping");

    gst_element_set_state(app->pipeline, GST_STATE_NULL);

    gst_object_unref(app->bus);
    g_main_loop_unref(app->gloop);
}

上面代码中,比较关键的地方是 appsrc 的 format 属性需要设置为 GST_FORMAT_TIME,如果不设置的话,视频和音频会无法同步,就是下面这几行代码:

...
g_object_set(app->shareScreenAppsrc, "format", GST_FORMAT_TIME, NULL);
...
g_object_set(app->sendAudioAppsrc, "format", GST_FORMAT_TIME, NULL);
...
g_object_set(app->receiveAudioAppsrc, "format", GST_FORMAT_TIME, NULL);

然后在需要启动录制的地方开启线程,启动 GStreamer 处理线程(因为项目比较复杂,这里只截取部分)

...
m_pAppSrcOption = new AppSrcOption();
m_pAppSrcOption->recordFileName = filePath;
m_pAppSrcOption->pushStreamUrl = m_pushStreamUrl;

m_pAppSrcOption->callbackData = this;
m_pAppSrcOption->sucCallback = OnLocalRecordSuc;
m_pAppSrcOption->errCallback = OnLocalRecordErr;

m_pLocalRecordProcessThread = new GStreamProcess();
m_pLocalRecordProcessThread->app = m_pAppSrcOption;

connect(m_pLocalRecordProcessThread, &GStreamProcess::finished, m_pLocalRecordProcessThread, &QObject::deleteLater);
connect(m_pLocalRecordProcessThread, &GStreamProcess::finished, this, &MainForm::OnLocalRecordClose);

m_pLocalRecordProcessThread->start();

m_pFillBlankAudioTimer->start(2000);
...

接下来注入视频帧数据。 这个项目在分享屏幕时,每一帧的视频数据会回调指定方法。在回调方法中,我们将数据传给管道中的 shareScreenAppsrc

void MainForm::localVideoHook(const LiveVideoFrame* video_frame)
{
    if (m_pAppSrcOption && m_pAppSrcOption->pipeline)
    {
        GstBuffer *buffer;
        guint8 *ptr;
        ptr = (guint8 *)g_malloc(video_frame->dataSize * sizeof(uint8));
        if (NULL == ptr)
        {
            qDebug("OnLocalVideo::malloc failed!");
        }
        else
        {
            memcpy(ptr, video_frame->data, video_frame->dataSize);
            buffer = gst_buffer_new_wrapped((void*)ptr, video_frame->dataSize);

            //设置时间戳
            GST_BUFFER_PTS(buffer) = gst_clock_get_time(m_pAppSrcOption->pipeline->clock) - m_pAppSrcOption->pipeline->base_time;

            GstFlowReturn ret;
            //注入视频帧数据
            g_signal_emit_by_name(m_pAppSrcOption->shareScreenAppsrc, "push-buffer", buffer, &ret);

            gst_buffer_unref(buffer);
        }
    }
}

上面代码中,设置时间戳的代码非常关键,如果没有时间戳,会导致管道中的 videorate 由于缺少时间戳信息而失败。
这里的时间戳取的是:当前管道时间 - 管道启动时的时间

然后以类似的方式注入音频帧数据,音频帧有两路,一路为老师的声音,一路为学生的声音。
老师的声音:

void MainForm::sendAudioHook(const iLiveAudioFrame* audio_frame)
{

    if (m_pAppSrcOption && m_pAppSrcOption->pipeline)
    {
        GstBuffer *buffer;

        guint8 *ptr;
        ptr = (guint8 *)g_malloc(audio_frame->dataSize * sizeof(uint8));
        if (NULL == ptr)
        {
            qDebug("OnSendAudioCallback::malloc failed!");
        }
        else
        {
            memcpy(ptr, audio_frame->data, audio_frame->dataSize);
            buffer = gst_buffer_new_wrapped((void*)ptr, audio_frame->dataSize);

            GstClockTime pts = gst_clock_get_time(m_pAppSrcOption->pipeline->clock) - m_pAppSrcOption->pipeline->base_time;
            GST_BUFFER_PTS(buffer) = pts;
            m_lastWriteSendAudioTime = pts;

            GST_DEBUG("feed buffer");

            GstFlowReturn ret;
            g_signal_emit_by_name(m_pAppSrcOption->sendAudioAppsrc, "push-buffer", buffer, &ret);

            gst_buffer_unref(buffer);
        }
    }
}

学生的声音:

void MainForm::receiveAudioHook(const iLiveAudioFrame* audio_frame)
{

    if (m_pAppSrcOption && m_pAppSrcOption->pipeline)
    {
        GstBuffer *buffer;

        guint8 *ptr;
        ptr = (guint8 *)g_malloc(audio_frame->dataSize * sizeof(uint8));
        if (NULL == ptr)
        {
            qDebug("receiveAudioHook::malloc failed!");
        }
        else
        {
            memcpy(ptr, audio_frame->data, audio_frame->dataSize);
            buffer = gst_buffer_new_wrapped((void*)ptr, audio_frame->dataSize);

            GstClockTime pts = gst_clock_get_time(m_pAppSrcOption->pipeline->clock) - m_pAppSrcOption->pipeline->base_time;
            GST_BUFFER_PTS(buffer) = pts;
            m_lastWriteReceiveAudioTime = pts;

            GST_DEBUG("feed buffer");

            GstFlowReturn ret;
            g_signal_emit_by_name(m_pAppSrcOption->receiveAudioAppsrc, "push-buffer", buffer, &ret);

            gst_buffer_unref(buffer);
        }
    }
}

在项目中还有一个特殊的地方,因为声音并不是一直有数据的,如果学生没有连麦或老师没有开启麦克风,是没有音频帧数据回调的,这会导致音频混流时因为缺少数据而一直等待,造成阻塞。

这里采取了一个变通的方法,就是设置一个定时器,每隔 2 秒注入一帧空白数据。上面代码中的 m_pFillBlankAudioTimer->start(2000); 就是用于启动这个定时器的。下面是注入空白帧的代码:

2018-03-15 updated: 这里每2秒注入一个空白帧会导致录制下来的视频出现声音突突声。后来改为增加了一个变量,在收到音频信号时记录上次插入音频帧的时间戳。每隔2秒打算插入空白帧时,检查时间戳距现在大于 5 秒才插入,否则不必插入。

//填充音频无信号时的空白声音,音频断流的话,会导致 adder 混流 block
void MainForm::OnFillBlankAudioTimer()
{
    if (m_pAppSrcOption && m_pAppSrcOption->pipeline && m_pAppSrcOption->receiveAudioAppsrc)
    {
        GstClockTime pts = gst_clock_get_time(m_pAppSrcOption->pipeline->clock) - m_pAppSrcOption->pipeline->base_time;

        if (GST_TIME_AS_SECONDS(pts - m_lastWriteReceiveAudioTime) > 5)
        {
            GstBuffer *buffer;
            guint size;
            GstFlowReturn ret;

            size = 3840;

            buffer = gst_buffer_new_allocate(NULL, size, NULL);

            //全部填入0x0
            gst_buffer_memset(buffer, 0, 0x0, size);

            GST_BUFFER_PTS(buffer) = pts;

            g_signal_emit_by_name(m_pAppSrcOption->receiveAudioAppsrc, "push-buffer", buffer, &ret);
            gst_buffer_unref(buffer);
        }
    }

    if (m_pAppSrcOption && m_pAppSrcOption->pipeline && m_pAppSrcOption->sendAudioAppsrc)
    {
        GstClockTime pts = gst_clock_get_time(m_pAppSrcOption->pipeline->clock) - m_pAppSrcOption->pipeline->base_time;

        if (GST_TIME_AS_SECONDS(pts - m_lastWriteSendAudioTime) > 5)
        {
            GstBuffer *buffer;
            guint size;
            GstFlowReturn ret;

            size = 3840;

            buffer = gst_buffer_new_allocate(NULL, size, NULL);

            //全部填入0x0
            gst_buffer_memset(buffer, 0, 0x0, size);

            GST_BUFFER_PTS(buffer) = pts;

            g_signal_emit_by_name(m_pAppSrcOption->sendAudioAppsrc, "push-buffer", buffer, &ret);
            gst_buffer_unref(buffer);
        }
    }
}

当需要结束混流和录制时,向管道中的各个 appsrc 发送 end-of-stream 消息,管道在处理完所有数据后,就会正常结束,关闭退出。

void MainForm::onBtnStopPushStream()
{
    QMessageBox::StandardButton ret = QMessageBox::question(this, FromBits("确认"), FromBits("是否要停止视频录制?(多次录制会产生多个视频文件,会影响回放的体验,应尽量避免多次录制)"));

    if (ret == QMessageBox::Yes)
    {
        stopPushStream();

        if (m_pAppSrcOption && m_pAppSrcOption->pipeline)
        {
            GstFlowReturn ret;
            g_signal_emit_by_name(m_pAppSrcOption->shareScreenAppsrc, "end-of-stream", &ret);
            g_signal_emit_by_name(m_pAppSrcOption->sendAudioAppsrc, "end-of-stream", &ret);
            g_signal_emit_by_name(m_pAppSrcOption->receiveAudioAppsrc, "end-of-stream", &ret);

            m_pFillBlankAudioTimer->stop();
        }

        m_pushStreamStatus = E_ChangingPushStream;
        setStatus(m_status);
    }
}

之前我们在启动线程时,用下面这句代码注册了事件,当线程结束时会调用 OnLocalRecordClose 方法,可以在这个方法中更改 UI 控件的状态和释放资源

...
connect(m_pLocalRecordProcessThread, &GStreamProcess::finished, this, &MainForm::OnLocalRecordClose);
...

因为第一次使用 GStreamer 进行开发,走了很多弯路,踩了很多坑。好在最后还是完成了需要的功能,直播录制和推流的效果还是不错的。

但这个方案还存在一个问题,就是rtmp 推流时如果失败,会导致整个管道出错停止,这个还需要设法解决。