Slide 1

Slide 1 text

Mac OS X 與 iOS 的 Audio API 楊維中 a.k.a zonble [email protected] Thursday, August 15,

Slide 2

Slide 2 text

與 Audio 有關的 API • AVFoundation • Audio Service • OpenAL • Audio Queue • Audio Unit • Audio Session • MPNowPlayingCenter • Background Task • Remote Control Events • … Thursday, August 15,

Slide 3

Slide 3 text

任務: 播放網站上的⾳音檔 ⽽而且,抓到了多少資料 就⾺馬上播放多少資料 Thursday, August 15,

Slide 4

Slide 4 text

這就不會是我們要的 • Audio Service:⽤用來處理簡短的⾳音效,例 如按到虛擬鍵盤上的喀嚓聲 • OpenAL:製作遊戲中的3D⾳音效,可以設 定⾳音效是從哪個⽅方向傳來的 Thursday, August 15,

Slide 5

Slide 5 text

AVAudioPlayer • ⾼高階的 Audio Player • iOS 2.2 • ⽀支援多種格式 • 可以從 file URL 或是 NSData 建⽴立 player 播放,但只能夠播放 local file • 適合製作遊戲背景⾳音樂 Thursday, August 15,

Slide 6

Slide 6 text

AVPlayer • ⾼高階的 Audio Player • iOS 4.0 • ⽀支援本機檔案與遠端檔案 • 適合網路廣播 •無法知道檔案總⻑⾧長度 Thursday, August 15,

Slide 7

Slide 7 text

我們的服務需要… • 可以直接讀多少資料就播放多少資料 • 可以⼀一邊下載⼀一邊 cache ⽇日後離線播放 • 知道載⼊入進度 • 檔案可能加密過,必須解密… • 對,有些公司就是會做這種事 Thursday, August 15,

Slide 8

Slide 8 text

我們需要了解底層 API CoreAudio Thursday, August 15,

Slide 9

Slide 9 text

因為太難了 所以先講簡單的… Thursday, August 15,

Slide 10

Slide 10 text

Audio Session • ⽤用來表⽰示⺫⽬目前的 Audio 屬於哪⼀一種類 • 設定 Audio Session Category • ⼀一般背景、Media Playback…etc • 然後呼叫 Audio Session Start • 有 C 與 Objective-C API Thursday, August 15,

Slide 11

Slide 11 text

Audio Session • 還要處理 Interrupt 與 Resume • 實作 delegate,更新 UI • Interrupt 的場合: • 別的 app 播放⾳音樂 • 來電 • 鬧鐘… Thursday, August 15,

Slide 12

Slide 12 text

回到 Core Audio Thursday, August 15,

Slide 13

Slide 13 text

說實在 這種 code 寫過⼀一次 不會想寫第⼆二次 Thursday, August 15,

Slide 14

Slide 14 text

請忘記 OO 忘記Objective-C Audio 串流就是快速處理⼀一堆 連續的 Binary 資料 Thursday, August 15,

Slide 15

Slide 15 text

基本觀念 • Audio 資料是連續的 binary 資料 • 資料中包含的是連續的sample/frame,⼀一 秒鐘會有 44100 個 sample • MP3/AAC等⺫⽬目前通⽤用的壓縮⾳音檔格式, 會將若干⼤大⼩小的frame變成⼀一個packet • 如果是VBR(變動碼率)的⾳音檔,每個 packet裡頭的資料量不⼀一樣⼤大 Thursday, August 15,

Slide 16

Slide 16 text

⼀一些術語 • Sample Rate:每秒鐘有多少 sample • Packet Size:每個 Packet 有多少 sample • Bit Rate:每秒鐘有多少 bit Thursday, August 15,

Slide 17

Slide 17 text

完整流程 • 建⽴立網路連線,從連線 callback 讀取資料 • 解密,並將解密過的資料放⼊入記憶體 • 建⽴立 Audio Queue 或是 Audio Unit Graph,收取系統通知需要下⼀一段資料的 callback • 提供資料 Thursday, August 15,

Slide 18

Slide 18 text

我們專注在 • 將資料讀⼊入記憶體 • Parse 出 packet 並保存 • 將資料餵給 Audio API • 註冊 callback • 在 callback 中回傳下⼀一段資料 • 資料要轉換成 Linear PCM 格式 Thursday, August 15,

Slide 19

Slide 19 text

Audio Unit 與 Audio Queue 兩組 API 的差別? • 在上⾴頁的流程中,最⼤大的差別在於Audio Queue API 不需要⼿手動將資料轉換成 Linear PCM 格式 • 不容易設定 • 可以直接對Linear PCM資料做⼿手腳… • Audio Unit 中可以在 Audio Graph 中增加 mixer 與 EQ effect node Thursday, August 15,

Slide 20

Slide 20 text

Packet ID3 data MP3 Header MP3 Data MP3 Header MP3 Data MP3 Header MP3 Data Packet Thursday, August 15,

Slide 21

Slide 21 text

辨識 MP3 header • MP3 Header 共 4 個 bytes • 其中前11個bit是sync word(都是1),看 到 sync word就可以知道是⼀一個packet的 開頭 • sync word之後描述這個packet的格式 • http://www.mp3-tech.org/programmer/ frame_header.html Thursday, August 15,

Slide 22

Slide 22 text

Sample Packet Parser def parse(content): i = 0 while i + 2 < len(content): frameSync = (content[i] << 8) | (content[i + 1] & (0x80 | 0x40 | 0x20)) if frameSync != 0xffe0: if foundFirstFrame: pass i += 1; continue if not foundFirstFrame: foundFirstFrame = True audioVersion = (content[i + 1] >> 3) & 0x03; layer = (content[i + 1] >> 1) & 0x03 hasCRC = not(content[i + 1] & 0x01) bitrateIndex = content[i + 2] >> 4; sampleRateIndex = content[i + 2] >> 2 & 0x03; bitrate = [0, 32000, 40000, 48000, 56000, 64000, 80000, 96000, 112000, 128000, 160000, 192000, 224000, 256000, 320000, 0][bitrateIndex]; hasPadding = not(not((content[i + 2] >> 1) & 0x01)) frameLength = 144 * bitrate / 44100 + \ (1 if hasPadding else 0) + \ (2 if hasCRC else 0) i += frameLength Thursday, August 15,

Slide 23

Slide 23 text

MPEG 4 Audio/Video MOOV MDAT Thursday, August 15,

Slide 24

Slide 24 text

CoreAudio 提供我們 ⼀一組 Audio Parser • ⽤用途是找出 Packet,並分析出檔案格式 • C API • iOS 2/Mac OS X 10.5 • 本機檔案可以呼叫 AudioFileOpenURL • 串流資料可以呼叫AudioFileStreamOpen Thursday, August 15,

Slide 25

Slide 25 text

AudioFileStreamOpen • AudioFileStreamOpen(self, ZBAudioFileStreamPropertyListener, ZBAudioFileStreamPacketsCallback, kAudioFileMP3Type, &audioFileStreamID); • self :傳遞⼀一個callback可以使⽤用的物件 • ZBAudioFileStreamPropertyListener 檔案格式 callback • ZBAudioFileStreamPacketsCallback parse 出 packet 的 callback • kAudioFileMP3Type 給 parser 的 hint • audioFileStreamID 產⽣生 audio file stream ID Thursday, August 15,

Slide 26

Slide 26 text

在記憶體中保存 packet? • 其實我們只要⽤用個簡單的 structure 就可以 保存 typedef struct { size_t length; // packet ⻑⾧長度 void *data; // packet 資料的指標 } Thursday, August 15,

Slide 27

Slide 27 text

保存檔案格式 void ZBAudioFileStreamPropertyListener(void * inClientData, AudioFileStreamID inAudioFileStream, AudioFileStreamPropertyID inPropertyID, UInt32 * ioFlags) { ZBSimplePlayer *self = (ZBSimplePlayer *)inClientData; if (inPropertyID == kAudioFileStreamProperty_DataFormat) { UInt32 dataSize = 0; OSStatus status = 0; AudioStreamBasicDescription audioStreamDescription; Boolean writable = false; status = AudioFileStreamGetPropertyInfo(inAudioFileStream, kAudioFileStreamProperty_DataFormat, &dataSize, &writable); status = AudioFileStreamGetProperty(inAudioFileStream, kAudioFileStreamProperty_DataFormat, &dataSize, &audioStreamDescription); // 然後把 audioStreamDescription 存起來 } } Thursday, August 15,

Slide 28

Slide 28 text

保存 Packet static void ZBAudioQueueOutputCallback(void * inUserData, AudioQueueRef inAQ,AudioQueueBufferRef inBuffer) { ZBSimplePlayer *self = (ZBSimplePlayer *)inClientData; for (int i = 0; i < inNumberPackets; ++i) { SInt64 packetStart = inPacketDescriptions[i].mStartOffset; UInt32 packetSize = inPacketDescriptions[i].mDataByteSize; assert(packetSize > 0); self->packetData[self->packetCount].length = (size_t)packetSize; self->packetData[self->packetCount].data = malloc(packetSize); memcpy(packetData[self->packetCount].data, inInputData + packetStart, packetSize); self->packetCount++; } } Thursday, August 15,

Slide 29

Slide 29 text

下⼀一步 • 為了要播放順利,我們會在有⼀一定數量 的 packet 之後,才會開始呼叫 Audio API 開始播放。資料不夠會產⽣生爆⾳音 • 等待 packet 的過程叫buffering…緩衝處理 • 資料換算成時間的⽅方式:packet 數量 * frames per packet / sample rate • 100 * 1152 / 44100 = 2.61... Thursday, August 15,

Slide 30

Slide 30 text

播放 Callback Callback Thursday, August 15,

Slide 31

Slide 31 text

Audio Queue 的播放 • 每次 callback,提供⼀一個新的buffer struct • buffer物件包含 • packet 數量 • 指向 packet 資料的指標 • packet格式 • 將 buffer 送⼊入 queue 中 Thursday, August 15,

Slide 32

Slide 32 text

AudioUnit 的播放 • Audio Unit API 會給你⼀一個 struct 的指, 標,叫做IOData,並傳⼊入⼀一定⼤大⼩小的 frame 數量 • 對IOData指定所需要⼤大⼩小的Linear PCM 資料 Thursday, August 15,

Slide 33

Slide 33 text

iOS 上的 Audio Unit • ⼀一般模式與畫⾯面鎖定時,要求的frame數 量不⼀一樣⼤大 • 平時要求1024,但畫⾯面鎖定時要求4096 • 主要考量是節電 • 如果資料量不夠,會停⽌止播放 Thursday, August 15,

Slide 34

Slide 34 text

Audio Queue Thursday, August 15,

Slide 35

Slide 35 text

建⽴立 Audio Queue OSStatus status = AudioQueueNewOutput(audioStreamBasicDes cription, ZBAudioQueueOutputCallback, self, CFRunLoopGetCurrent(), kCFRunLoopCommonModes, 0, &outputQueue); assert(status == noErr); Thursday, August 15,

Slide 36

Slide 36 text

Enqueue Data AudioQueueBufferRef buffer; status = AudioQueueAllocateBuffer(outputQueue, totalSize, &buffer); assert(status == noErr); buffer->mAudioDataByteSize = totalSize; buffer->mUserData = self; AudioStreamPacketDescription *packetDescs = calloc(inPacketCount, sizeof(AudioStreamPacketDescription)); totalSize = 0; for (index = 0 ; index < inPacketCount ; index++) { size_t readIndex = index + readHead; memcpy(buffer->mAudioData + totalSize, packetData[readIndex].data, packetData[readIndex].length); AudioStreamPacketDescription description; description.mStartOffset = totalSize; description.mDataByteSize = (UInt32)packetData[readIndex].length; description.mVariableFramesInPacket = 0; totalSize += packetData[readIndex].length; memcpy(&(packetDescs[index]), &description, sizeof(AudioStreamPacketDescription)); } status = AudioQueueEnqueueBuffer(outputQueue, buffer, (UInt32)inPacketCount, packetDescs); free(packetDescs); Thursday, August 15,

Slide 37

Slide 37 text

Audio Unit Thursday, August 15,

Slide 38

Slide 38 text

Audio Unit Output Node Effect Node Mixer Node Render Callback Audio Unit Graph Mixer Unit Effect Unit Output Unit Get Info Get Info Get Info Thursday, August 15,

Slide 39

Slide 39 text

建⽴立 audio converter AudioConverterRef converter; AudioStreamBasicDescription fromFormat=…⋯; AudioStreamBasicDescription destFormat=…⋯; AudioConverterNew(&fromFormat, &destFormat, &converter); Thursday, August 15,

Slide 40

Slide 40 text

使⽤用 audio converter AudioBufferList *list; UInt32 packetSize = 1024; AudioConverterFillComplexBuffer(conve rter, ZBPlayerConverterFiller, self, &packetSize, list, NULL); Thursday, August 15,

Slide 41

Slide 41 text

audio converter callback OSStatus ZBPlayerConverterFiller (AudioConverterRef inAudioConverter, UInt32* ioNumberDataPackets, AudioBufferList* ioData, AudioStreamPacketDescription** outDataPacketDescription, void* inUserData) { ZBSimpleAUPlayer *self = (ZBSimpleAUPlayer *)inUserData; *ioNumberDataPackets = 1; static AudioStreamPacketDescription aspdesc; ioData->mNumberBuffers = 1; void *data = self->packetData[readHead].data; UInt32 length = self->packetData[readHead].length; ioData->mBuffers[0].mData = data; ioData->mBuffers[0].mDataByteSize = length; *outDataPacketDescription = &aspdesc; aspdesc.mDataByteSize = length; aspdesc.mStartOffset = 0; aspdesc.mVariableFramesInPacket = 1; readHead++; return noErr; } Thursday, August 15,

Slide 42

Slide 42 text

NewAUGraph(&audioGraph); // 建立 audio grapg AudioComponentDescription cdesc; bzero(&cdesc, sizeof(AudioComponentDescription)); cdesc.componentType = kAudioUnitType_Output; cdesc.componentSubType = kAudioUnitSubType_DefaultOutput; cdesc.componentManufacturer = kAudioUnitManufacturer_Apple; cdesc.componentFlags = 0; cdesc.componentFlagsMask = 0; AUGraphAddNode(audioGraph, &cdesc, &outputNode); // 建立 output node AUGraphOpen(audioGraph); AUGraphNodeInfo(audioGraph, outputNode, &cdesc, &outputUnit); // 建立 output unit AudioStreamBasicDescription destFormat = LFPCMStreamDescription(); // 設定輸出格式 AudioUnitSetProperty(outputUnit, kAudioUnitProperty_StreamFormat, kAudioUnitScope_Input, 0, &destFormat, sizeof(destFormat)); AURenderCallbackStruct callbackStruct; callbackStruct.inputProc = ZBPlayerAURenderCallback; callbackStruct.inputProcRefCon = self; AudioUnitSetProperty(outputUnit, kAudioUnitProperty_SetRenderCallback, kAudioUnitScope_Input, 0, &callbackStruct, sizeof(callbackStruct)); // 建立 render callback AUGraphInitialize(audioGraph); // Init audio graph CAShow(audioGraph); // 開始 audio graph Thursday, August 15,

Slide 43

Slide 43 text

播放 • AUGraphStart(audioGraph); • AUGraphStop(audioGraph); Thursday, August 15,

Slide 44

Slide 44 text

Sample Code • https://github.com/zonble/ZBSimplePlayer Thursday, August 15,

Slide 45

Slide 45 text

因為 Audio Unit API 需 要直接使⽤用 Linear PCM 資料,我們也可 以直接修改資料產⽣生 效果 Thursday, August 15,

Slide 46

Slide 46 text

Thanks! Thursday, August 15,