[feat] 为rtmp pub推流添加静音AAC音频(可动态检测是否需要添加；配置文件中可开启或关闭这个功能) (#56)

.gitignore

@@ -1,3 +1,4 @@
+delay.txt
 profile.out
 coverage.html
 *.aac

conf/lalserver.conf.json

@@ -1,11 +1,13 @@
 {
   "# doc of config": "https://pengrl.com/lal/#/ConfigBrief",
-  "conf_version": "v0.2.2",
+  "conf_version": "v0.2.3",
   "rtmp": {
     "enable": true,
     "addr": ":1935",
     "gop_num": 0,
-    "merge_write_size": 0
+    "merge_write_size": 0,
+    "add_dummy_audio_enable": false,
+    "add_dummy_audio_wait_audio_ms": 150
   },
   "default_http": {
     "http_listen_addr": ":8080",

pkg/aac/seqheader.go

@@ -48,8 +48,8 @@ func MakeAudioDataSeqHeaderWithAsc(asc []byte) (out []byte, err error) {
 		return nil, ErrAac
 	}
 
+	// 注意，前两个字节是SequenceHeaderContext，后面跟着asc
 	out = make([]byte, 2+len(asc))
-	// <spec-video_file_format_spec_v10.pdf>, <Audio tags, AUDIODATA>, <page 10/48>
 	out[0] = 0xaf
 	out[1] = 0
 	copy(out[2:], asc)

pkg/logic/config.go

@@ -54,6 +54,8 @@ type RtmpConfig struct {
 	Addr                     string `json:"addr"`
 	GopNum                   int    `json:"gop_num"`
 	MergeWriteSize           int    `json:"merge_write_size"`
+	AddDummyAudioEnable      bool   `json:"add_dummy_audio_enable"`
+	AddDummyAudioWaitAudioMs int    `json:"add_dummy_audio_wait_audio_ms"`
 }
 
 type DefaultHttpConfig struct {

pkg/logic/dummy_audio_filter.go

@@ -0,0 +1,211 @@
+// Copyright 2021, Chef.  All rights reserved.
+// https://github.com/q191201771/lal
+//
+// Use of this source code is governed by a MIT-style license
+// that can be found in the License file.
+//
+// Author: Chef (191201771@qq.com)
+
+package logic
+
+import (
+	"github.com/q191201771/lal/pkg/base"
+	"github.com/q191201771/lal/pkg/rtmp"
+	"github.com/q191201771/naza/pkg/nazalog"
+	"math"
+)
+
+const dummyAudioFilterStageAnalysis = 1
+const dummyAudioFilterStageNormal = 2
+const dummyAudioFilterStageDummy = 3
+
+type DummyAudioFilter struct {
+	uk          string
+	waitAudioMs int
+	onPop       rtmp.OnReadRtmpAvMsg
+
+	stage           int
+	earlyStageQueue []base.RtmpMsg
+	firstVideoTs    uint32
+	prevAudioTs     uint32
+
+	audioCount int
+}
+
+// NewDummyAudioFilter 检测输入的rtmp流中是否有音频，如果有，则原样返回；如果没有，则制造静音音频数据叠加在rtmp流里面
+//
+// @param waitAudioMs 等待音频数据时间，如果超出这个时间还没有接收到音频数据，则开始制造静音数据
+// @param onPop       注意，所有回调都发生在输入函数调用中
+//
+func NewDummyAudioFilter(uk string, waitAudioMs int, onPop rtmp.OnReadRtmpAvMsg) *DummyAudioFilter {
+	return &DummyAudioFilter{
+		uk:           uk,
+		waitAudioMs:  waitAudioMs,
+		onPop:        onPop,
+		stage:        dummyAudioFilterStageAnalysis,
+		firstVideoTs: math.MaxUint32,
+		prevAudioTs:  math.MaxUint32,
+	}
+}
+
+func (filter *DummyAudioFilter) OnReadRtmpAvMsg(msg base.RtmpMsg) {
+	filter.Feed(msg)
+}
+
+func (filter *DummyAudioFilter) Feed(msg base.RtmpMsg) {
+	//nazalog.Debugf("trace_DummyAudioFilter, push. header=%+v, payload=%s, stage=%d", msg.Header, hex.EncodeToString(nazastring.SubSliceSafety(msg.Payload, 8)), filter.stage)
+	switch filter.stage {
+	case dummyAudioFilterStageAnalysis:
+		filter.handleAnalysisStage(msg)
+	case dummyAudioFilterStageNormal:
+		filter.handleNormalStage(msg)
+	case dummyAudioFilterStageDummy:
+		filter.handleDummyStage(msg)
+	}
+}
+
+// 初始阶段，分析是否存在音频
+func (filter *DummyAudioFilter) handleAnalysisStage(msg base.RtmpMsg) {
+	switch msg.Header.MsgTypeId {
+	case base.RtmpTypeIdMetadata:
+		// metadata直接入队列
+		filter.cache(msg)
+	case base.RtmpTypeIdAudio:
+		// 原始流中存在音频，将所有缓存数据出队列，进入normal stage
+		for i := range filter.earlyStageQueue {
+			filter.onPopProxy(filter.earlyStageQueue[i])
+		}
+		filter.onPopProxy(msg)
+
+		filter.stage = dummyAudioFilterStageNormal
+	case base.RtmpTypeIdVideo:
+		// 分析视频数据累计时长是否达到阈值
+
+		// 注意，为了避免seq header的时间戳和视频帧不是线性的（0或其他特殊的值）我们直接入队列并跳过
+		if msg.IsVideoKeySeqHeader() {
+			filter.cache(msg)
+			return
+		}
+
+		// 记录首个视频帧的时间戳
+		if filter.firstVideoTs == math.MaxUint32 {
+			filter.cache(msg)
+			filter.firstVideoTs = msg.Header.TimestampAbs
+			return
+		}
+
+		// 没有达到阈值
+		if msg.Header.TimestampAbs-filter.firstVideoTs < uint32(filter.waitAudioMs) {
+			filter.cache(msg)
+			return
+		}
+
+		// 达到阈值
+		nazalog.Debugf("[%s] start make dummy audio.", filter.uk)
+		filter.stage = dummyAudioFilterStageDummy
+		for i := range filter.earlyStageQueue {
+			filter.handleDummyStage(filter.earlyStageQueue[i])
+		}
+		filter.clearCache()
+		filter.handleDummyStage(msg)
+	}
+}
+
+// 原始流中存在音频
+func (filter *DummyAudioFilter) handleNormalStage(msg base.RtmpMsg) {
+	filter.onPopProxy(msg)
+}
+
+// 原始流中不存在音频
+func (filter *DummyAudioFilter) handleDummyStage(msg base.RtmpMsg) {
+	if msg.Header.MsgTypeId == base.RtmpTypeIdAudio {
+		// 由于我们已经开始制造静音包了，静音包的编码参数可能会和实际音频参数不一致，所以我们只能过滤掉原始的音频数据了
+		nazalog.Warnf("[%s] recv audio but we are making dummy audio.", filter.uk)
+		return
+	}
+
+	if msg.Header.MsgTypeId == base.RtmpTypeIdMetadata {
+		filter.onPopProxy(msg)
+		return
+	}
+
+	if msg.IsVideoKeySeqHeader() {
+		// TODO(chef): 这里的时间戳可以考虑减1，但是注意处理一些边界条件
+		ats := msg.Header.TimestampAbs
+		amsg := filter.makeAudioSeqHeader(ats)
+		filter.onPopProxy(amsg)
+		filter.onPopProxy(msg)
+		return
+	}
+
+	if filter.prevAudioTs == math.MaxUint32 {
+		ats := msg.Header.TimestampAbs
+		amsg := filter.makeOneAudio(ats)
+		filter.onPopProxy(amsg)
+		filter.onPopProxy(msg)
+		filter.prevAudioTs = ats
+	} else {
+		for {
+			ats := filter.prevAudioTs + filter.calcAudioDurationMs()
+			if ats > msg.Header.TimestampAbs {
+				break
+			}
+			amsg := filter.makeOneAudio(ats)
+			filter.onPopProxy(amsg)
+			filter.prevAudioTs = ats
+		}
+		filter.onPopProxy(msg)
+	}
+}
+
+func (filter *DummyAudioFilter) cache(msg base.RtmpMsg) {
+	filter.earlyStageQueue = append(filter.earlyStageQueue, msg.Clone())
+}
+
+func (filter *DummyAudioFilter) clearCache() {
+	filter.earlyStageQueue = nil
+}
+
+func (filter *DummyAudioFilter) onPopProxy(msg base.RtmpMsg) {
+	//nazalog.Debugf("trace_DummyAudioFilter, onPopProxy. header=%+v, payload=%s, stage=%d", msg.Header, hex.EncodeToString(nazastring.SubSliceSafety(msg.Payload, 8)), filter.stage)
+	if filter.onPop != nil {
+		filter.onPop(msg)
+	}
+}
+
+func (filter *DummyAudioFilter) makeAudioSeqHeader(ts uint32) base.RtmpMsg {
+	// aac (LC), 48000 Hz, stereo, fltp
+	return base.RtmpMsg{
+		Header: base.RtmpHeader{
+			Csid:         rtmp.CsidAudio,
+			MsgLen:       4,
+			MsgTypeId:    base.RtmpTypeIdAudio,
+			MsgStreamId:  rtmp.Msid1,
+			TimestampAbs: ts,
+		},
+		Payload: []byte{0xaf, 0x00, 0x11, 0x90},
+	}
+}
+
+func (filter *DummyAudioFilter) makeOneAudio(ts uint32) base.RtmpMsg {
+	filter.audioCount++
+	return base.RtmpMsg{
+		Header: base.RtmpHeader{
+			Csid:         rtmp.CsidAudio,
+			MsgLen:       8,
+			MsgTypeId:    base.RtmpTypeIdAudio,
+			MsgStreamId:  rtmp.Msid1,
+			TimestampAbs: ts,
+		},
+		// 注意，前面2字节是seq header头部信息，后面6个字节是AAC静音包
+		Payload: []byte{0xaf, 0x01, 0x21, 0x10, 0x04, 0x60, 0x8c, 0x1c},
+	}
+}
+
+func (filter *DummyAudioFilter) calcAudioDurationMs() uint32 {
+	v := filter.audioCount % 3
+	if v == 1 || v == 2 {
+		return 21
+	}
+	return 22
+}

pkg/logic/dummy_audio_filter_test.go

@@ -0,0 +1,129 @@
+package logic_test
+
+import (
+	"encoding/hex"
+	"github.com/q191201771/lal/pkg/base"
+	"github.com/q191201771/lal/pkg/logic"
+	"github.com/q191201771/naza/pkg/assert"
+	"github.com/q191201771/naza/pkg/nazalog"
+	"strconv"
+	"strings"
+	"testing"
+)
+
+func TestDummyAudioFilter(t *testing.T) {
+	// case1 一个音视频都有的流
+	{
+		in := []base.RtmpMsg{
+			helperUnpackRtmpMsg("header={Csid:4 MsgLen:378 MsgTypeId:18 MsgStreamId:1 TimestampAbs:0}, payload=02000d40"),
+			helperUnpackRtmpMsg("header={Csid:6 MsgLen:48 MsgTypeId:9 MsgStreamId:1 TimestampAbs:0}, payload=17000000"),
+			helperUnpackRtmpMsg("header={Csid:4 MsgLen:7 MsgTypeId:8 MsgStreamId:1 TimestampAbs:0}, payload=af001210"),
+			helperUnpackRtmpMsg("header={Csid:4 MsgLen:26 MsgTypeId:8 MsgStreamId:1 TimestampAbs:0}, payload=af01de04"),
+			helperUnpackRtmpMsg("header={Csid:6 MsgLen:1170 MsgTypeId:9 MsgStreamId:1 TimestampAbs:23}, payload=17010000"),
+			helperUnpackRtmpMsg("header={Csid:4 MsgLen:8 MsgTypeId:8 MsgStreamId:1 TimestampAbs:23}, payload=af012110"),
+			helperUnpackRtmpMsg("header={Csid:4 MsgLen:8 MsgTypeId:8 MsgStreamId:1 TimestampAbs:46}, payload=af012120"),
+			helperUnpackRtmpMsg("header={Csid:4 MsgLen:849 MsgTypeId:8 MsgStreamId:1 TimestampAbs:69}, payload=af01214c"),
+			helperUnpackRtmpMsg("header={Csid:6 MsgLen:372 MsgTypeId:9 MsgStreamId:1 TimestampAbs:90}, payload=27010000"),
+		}
+		var out []base.RtmpMsg
+		filter := logic.NewDummyAudioFilter("test1", 150, func(msg base.RtmpMsg) {
+			out = append(out, msg)
+		})
+		//filter.Feed(helperUnpackRtmpMsg(""))
+		for i := 0; i <= 1; i++ {
+			filter.Feed(in[i])
+			assert.Equal(t, nil, out)
+		}
+		for i := 2; i < len(in); i++ {
+			filter.Feed(in[i])
+			assert.Equal(t, in[:i+1], out)
+		}
+	}
+
+	// case2 一个只有视频的流
+	{
+		in := []base.RtmpMsg{
+			helperUnpackRtmpMsg("header={Csid:4 MsgLen:269 MsgTypeId:18 MsgStreamId:1 TimestampAbs:0}, payload=02000d4073657444"),
+			helperUnpackRtmpMsg("header={Csid:6 MsgLen:48 MsgTypeId:9 MsgStreamId:1 TimestampAbs:0}, payload=1700000000016400"),
+			helperUnpackRtmpMsg("header={Csid:6 MsgLen:1170 MsgTypeId:9 MsgStreamId:1 TimestampAbs:23}, payload=1701000000000002"),
+			helperUnpackRtmpMsg("header={Csid:6 MsgLen:372 MsgTypeId:9 MsgStreamId:1 TimestampAbs:90}, payload=2701000000000001"),
+			helperUnpackRtmpMsg("header={Csid:6 MsgLen:1226 MsgTypeId:9 MsgStreamId:1 TimestampAbs:156}, payload=2701000000000004"),
+			helperUnpackRtmpMsg("header={Csid:6 MsgLen:1541 MsgTypeId:9 MsgStreamId:1 TimestampAbs:223}, payload=2701000000000005"),
+			helperUnpackRtmpMsg("header={Csid:6 MsgLen:1931 MsgTypeId:9 MsgStreamId:1 TimestampAbs:290}, payload=2701000000000005"),
+		}
+		var out []base.RtmpMsg
+		filter := logic.NewDummyAudioFilter("test1", 150, func(msg base.RtmpMsg) {
+			out = append(out, msg)
+		})
+		for i := 0; i <= 4; i++ {
+			filter.Feed(in[i])
+			assert.Equal(t, nil, out)
+		}
+		filter.Feed(in[5])
+		assert.Equal(t, 17, len(out))
+		assert.Equal(t, in[0], out[0])
+		assert.Equal(t, helperUnpackRtmpMsg("header={Csid:6 MsgLen:4 MsgTypeId:8 MsgStreamId:1 TimestampAbs:0}, payload=af001190"), out[1])
+		assert.Equal(t, in[1], out[2])
+		assert.Equal(t, helperUnpackRtmpMsg("header={Csid:6 MsgLen:8 MsgTypeId:8 MsgStreamId:1 TimestampAbs:215}, payload=af01211004608c1c"), out[15])
+		assert.Equal(t, in[5], out[16])
+
+		filter.Feed(in[6])
+		assert.Equal(t, 21, len(out))
+		assert.Equal(t, helperUnpackRtmpMsg("header={Csid:6 MsgLen:8 MsgTypeId:8 MsgStreamId:1 TimestampAbs:236}, payload=af01211004608c1c"), out[17])
+		assert.Equal(t, in[6], out[20])
+	}
+
+	// case3 一个只有音频的流
+	{
+		in := []base.RtmpMsg{
+			helperUnpackRtmpMsg("header={Csid:4 MsgLen:278 MsgTypeId:18 MsgStreamId:1 TimestampAbs:0}, payload=02000d4073657444"),
+			helperUnpackRtmpMsg("header={Csid:4 MsgLen:7 MsgTypeId:8 MsgStreamId:1 TimestampAbs:0}, payload=af00121056e500"),
+			helperUnpackRtmpMsg("header={Csid:4 MsgLen:26 MsgTypeId:8 MsgStreamId:1 TimestampAbs:0}, payload=af01de04004c6176"),
+			helperUnpackRtmpMsg("header={Csid:4 MsgLen:8 MsgTypeId:8 MsgStreamId:1 TimestampAbs:23}, payload=af01211004608c1c"),
+		}
+		var out []base.RtmpMsg
+		filter := logic.NewDummyAudioFilter("test1", 150, func(msg base.RtmpMsg) {
+			out = append(out, msg)
+		})
+		filter.Feed(in[0])
+		assert.Equal(t, nil, out)
+		for i := 1; i <= 3; i++ {
+			filter.Feed(in[i])
+			assert.Equal(t, in[:i+1], out)
+		}
+	}
+}
+
+// @param logstr e.g. "header={Csid:4 MsgLen:378 MsgTypeId:18 MsgStreamId:1 TimestampAbs:0}"
+///
+func helperUnpackRtmpMsg(logstr string) base.RtmpMsg {
+	var fetchItemFn = func(str string, prefix string, suffix string) string {
+		b := strings.Index(str, prefix)
+		if suffix == "" {
+			return str[b+len(prefix):]
+		}
+		e := strings.Index(str[b:], suffix)
+		return str[b+len(prefix) : b+e]
+	}
+	var fetchIntItemFn = func(str string, prefix string, suffix string) int {
+		ret, err := strconv.Atoi(fetchItemFn(str, prefix, suffix))
+		nazalog.Assert(nil, err)
+		return ret
+	}
+
+	var header base.RtmpHeader
+	header.Csid = fetchIntItemFn(logstr, "Csid:", " ")
+	header.MsgLen = uint32(fetchIntItemFn(logstr, "MsgLen:", " "))
+	header.MsgTypeId = uint8(fetchIntItemFn(logstr, "MsgTypeId:", " "))
+	header.MsgStreamId = fetchIntItemFn(logstr, "MsgStreamId:", " ")
+	header.TimestampAbs = uint32(fetchIntItemFn(logstr, "TimestampAbs:", "}"))
+
+	hexStr := fetchItemFn(logstr, "payload=", "")
+	payload, err := hex.DecodeString(hexStr)
+	nazalog.Assert(nil, err)
+
+	return base.RtmpMsg{
+		Header:  header,
+		Payload: payload,
+	}
+}

pkg/logic/group.go

@@ -77,6 +77,8 @@ type Group struct {
 	// rtmp pub/pull使用
 	rtmpGopCache    *GopCache
 	httpflvGopCache *GopCache
+	// rtmp pub使用
+	dummyAudioFilter *DummyAudioFilter
 	// rtmp sub使用
 	rtmpBufWriter base.IBufWriter // TODO(chef): 后面可以在业务层加一个定时Flush
 	// mpegts使用
@@ -317,7 +319,14 @@ func (group *Group) AddRtmpPubSession(session *rtmp.ServerSession) bool {
 		)
 	}
 
+	// TODO(chef): 为rtmp pull以及rtsp也添加叠加静音音频的功能
+	if config.RtmpConfig.AddDummyAudioEnable {
+		// TODO(chef): 从整体控制和锁关系来说，应该让pub的数据回调到group中进锁后再让数据流入filter
+		group.dummyAudioFilter = NewDummyAudioFilter(group.UniqueKey, config.RtmpConfig.AddDummyAudioWaitAudioMs, group.OnReadRtmpAvMsg)
+		session.SetPubSessionObserver(group.dummyAudioFilter)
+	} else {
 		session.SetPubSessionObserver(group)
+	}
 
 	return true
 }
@@ -640,6 +649,7 @@ func (group *Group) delRtmpPubSession(session *rtmp.ServerSession) {
 
 	group.rtmpPubSession = nil
 	group.rtmp2RtspRemuxer = nil
+	group.dummyAudioFilter = nil
 	group.delIn()
 }
 
@@ -1041,7 +1051,7 @@ func (group *Group) broadcastByRtmpMsg(msg base.RtmpMsg) {
 				// GOP缓存中肯定包含了关键帧
 				session.ShouldWaitVideoKeyFrame = false
 
-				nazalog.Debugf("[%s] [%s] write gop cahe. gop num=%d", group.UniqueKey, session.UniqueKey(), gopCount)
+				nazalog.Debugf("[%s] [%s] write gop cache. gop num=%d", group.UniqueKey, session.UniqueKey(), gopCount)
 			}
 			for i := 0; i < gopCount; i++ {
 				for _, item := range group.rtmpGopCache.GetGopDataAt(i) {