張智星 (Roger Jang) 台大資訊系 多媒體檢索實驗室 CSIE/NTU, MIR Lab MATLAB 程式設計入門篇 音訊讀寫、錄製與播放 (Audio Reading, Writing, Recording, and Playback) 張智星 (Roger Jang) 台大資訊系 多媒體檢索實驗室 CSIE/NTU, MIR Lab

音訊的基本介紹 聲音訊號（Audio Signals）簡稱音訊，泛指由人耳聽到的各種聲音的訊號。 音訊的基本聲學特徵 音量（Volume）：聲音的大小稱為音量，又稱為力度、強度（Intensity）或是能量（Energy）。音量越大，代表音訊波形的震幅越大。 音高（Pitch）：聲音的基本頻率（Fundamental Frequency）越高，代表音高越高（例如女高音的歌聲）；反之，聲音的基本頻率越低，代表音高越低（例如男低音的歌聲）。 音色（Timbre）：音訊波形在每個週期內的變化，就形成了此音訊的音色。不同的音色即代表不同的音訊內容，例如不同的字有不同的發音，或是不同的歌手有不同的特色，這些都是由於音色不同而產生。 Demo via CoolEdit

Basics about Audio Signals Audio signals: signals that are audible to human Basic perceptible acoustic features of speech Volume (音量): the amplitude of audio signals Also known as intensity, or energy. Pitch (音高): Fundamental frequency (the number of fundamental periods in a second) in audio signals. Usually males have a lower pitch while females have a higher one Timbre (音色): Waveform inside a fundamental period. Different vowels have different timbres Different singers also have different timbres. Demo via CoolEdit Quiz!

Timbre: Waveform within an FP Time-domain Features Time-domain audio features presented in a frame (analysis window of 20-40 ms) Fundamental period Intensity Timbre: Waveform within an FP

Frequency-domain Features Frequency-domain audio features in a frame Energy: Sum of power spectrum Pitch: Distance between harmonics Timbre: Smoothed spectrum Second formant F2 First formant F1 Pitch freq Energy

音檔的讀取、寫檔與播放 使用audioread讀取wav檔案，畫出音訊的波形並播放此音訊： audioRead01.m [y, fs]=audioread('welcome.wav'); sound(y, fs); % 播放此音訊 time=(1:length(y))/fs; % 時間軸的向量 plot(time, y); % 畫出時間軸上的波形 放大波形後即可看到基本週期！

Read, Write and Playback Use “audioread” to read a .wav file, plot its waveform and play the sound. audioRead01.m [y, fs]=audioread('welcome.wav'); sound(y, fs); % Playback time=(1:length(y))/fs; % Time vector plot(time, y); % Waveform display Enlarge to see fundamental periods!

Read Metadata from .wav Files 中介資料 Read Metadata from .wav Files Reading metadata info=audioInfo('file'); Different types of audio files may return different fields of info. Two types of reading data from audio files Read audio signals y= audioread('file') Read metadata info=audioinfo('file') Metadata of a .wav file audioInfo01.m fileName='flanger.wav'; info=audioinfo(fileName); fprintf('檔案名稱 = %s\n', info.Filename); fprintf('壓縮方式 = %s\n', info.CompressionMethod); fprintf('通道個數 = %g 個\n', info.NumChannels); fprintf('取樣頻率 = %g Hz\n', info.SampleRate); fprintf('取樣點總個數 = %g 個\n', info.TotalSamples); fprintf('音訊長度 = %g 秒\n', info.Duration); fprintf('取樣點解析度 = %g 位元/取樣點\n', info.BitsPerSample);

Metadata of Other Audio Files *.aif audioInfo02.m *.mp3 audioInfo03.m fileName='whale.aif'; info=audioinfo(fileName); disp(info); fileName='youAtLeast.mp3'; info=audioinfo(fileName); disp(info); Filename: 'D:\users\... CompressionMethod: 'Uncompressed' NumChannels: 1 SampleRate: 2000 TotalSamples: 4000 Duration: 2 Title: [] Comment: [] Artist: [] BitsPerSample: 16 Filename: 'D:\users\... CompressionMethod: 'MP3' NumChannels: 2 SampleRate: 44100 TotalSamples: 317953 Duration: 7.2098 Title: '02_至少還有你' Comment: [] Artist: '林憶蓮' BitRate: 128

Scaling of Audio Signals by “audioread” Internal data types of audio signals in a file (音訊檔案內部儲存方式) 8 bits  uint8, [0, 28-1] 16 bits  int16, [-215, 215-1] MATLAB’s method to scale raw audio signals to the range [-1, 1] 8 bits  (y-128)/128 16 bits  y/32768 Verification of MATLABs’ scaling audioRead03.m difference = 0 fileName='welcome.wav'; [y, fs]=audioread(fileName); info=audioinfo(fileName); nbits=info.BitsPerSample; % y0 是原先儲存在音訊檔案中的值 y0=y*(2^nbits/2)+(2^nbits/2); difference=sum(abs(y0-round(y0))) Original audio signals in integer

讀取雙聲道檔案 audioread 可以讀取雙聲道或立體聲（Stereo）的音檔，此時傳回的變數具有兩直行，每一直行代表一個聲道的音訊。 audioRead04.m fileName='flanger.wav'; [y, fs]=audioread(fileName); % 讀取音訊檔 sound(y, fs); % 播放音訊 left=y(:,1); % 左聲道音訊 right=y(:,2); % 右聲道音訊 subplot(2,1,1), plot((1:length(left))/fs, left); subplot(2,1,2), plot((1:length(right))/fs, right); 聲音會在左右喇叭游移！

Read Stereo Audio Files “audioread” can also read stereo audio files. The returned variable has two columns representing two channels of audio signals. audioRead04.m fileName='flanger.wav'; [y, fs]=audioread(fileName); % Read sound(y, fs); % Playback left=y(:,1); % Left channel right=y(:,2); % Right channel subplot(2,1,1), plot((1:length(left))/fs, left); subplot(2,1,2), plot((1:length(right))/fs, right); Moving sound source between two speakers!

讀取部分音檔 如果音檔很大，無法一次讀入記憶體，我們也可以使用 audioread 來讀出音檔的一部份，例如： audioRead05.m [y,fs]=audioread('welcome.wav', [4001 5000]); % 讀取第4001至5000點 figure; plot(y) 可看到明顯的基本週期！

Read a Portion Only If the audio file is too big, we can read a portion from the whole file: audioRead05.m [y,fs]=audioread(‘welcome.wav’, [4001 5000]); % Read data points 4001 to 5000 figure; plot(y) Obvious fundamental periods!

聲音訊號的播放 一旦我們可以讀入音訊檔案，就可以對聲音訊號進行各種處理，例如增大或減小音量、提高或降低音高、消除雜訊等。 要確認處理後的聲音訊號是否符合所需，就要能夠把音訊直接透過接到電腦的喇叭播放出來，本節就是要介紹如何使用 MATLAB 來進行音訊的播放。

Playback of Audio Signals Once we have read audio signals, we can perform all kinds of processing, such as volume modification, pitch scaling, noise reduction, etc. To verify the result, you need to play the audio via speakers, as shown in the following examples.

播放聲音 (1/2) 一旦 MATLAB 讀入音訊資料，並將之設定成工作空間中的變數後，我們就可以使用 sound 指令來直接播放此變數。 播放單一聲音 audioPlay01.m 同時播放兩種聲音 audioPlay02.m load handel.mat % 載入音訊 sound(y, Fs); % 播放音訊 [y, fs]=audioread('welcome.wav'); % 載入音訊 sound(5*y, fs); % 播放音訊 load handel.mat % 載入音訊 sound(y, Fs); % 播放音訊 sound的播放模式為「非同步播放」！

Playback (1/2) We can use “sound” to play audio signals that has been read and stored as a variable in MATLAB’s workspace. Audio playback audioPlay01.m Simultaneous playback audioPlay02.m load handel.mat % Load audio sound(y, Fs); % Playback [y, fs]=audioread('welcome.wav'); % Read audio sound(5*y, fs); % Playback load handel.mat % Load audio sound(y, Fs); % Playback Playback mode of “sound” is asynchronous!

播放聲音 (2/2) 若要控制聲音的播放模式，則必須採用功能較為強大的指令： 播放單一聲音 循序播放兩種聲音 audioPlay03.m audioplayer play playblocking 播放單一聲音 audioPlay03.m 循序播放兩種聲音 audioPlay04.m load handel.mat % 載入音訊 p=audioplayer(y, Fs);% 產生播放物件 play(p); % 播放音訊 [y, fs]=audioread(‘welcome.wav’); % 讀入音訊 p=audioplayer(y, fs); % 產生播放物件 playblocking(p); % 播放音訊 load handel.mat % 載入音訊 p=audioplayer(y, Fs); % 產生播放物件 playblocking 的播放模式為「同步播放」！

Playback (2/2) If you want to control the playback mode, you need to invoke other commands: audioplayer play playblocking Single playback audioPlay03.m Sequential playback audioPlay04.m load handel.mat % Load audio p=audioplayer(y, Fs);% Player object play(p); % Playback [y, fs]=audioread(‘welcome.wav’); % Read audio p=audioplayer(y, fs); % Player object playblocking(p); % Playback load handel.mat % Load audio p=audioplayer(y, Fs); % Player object Playback mode of “playblocking” is synchronous!

改變音訊的震幅 我們在第一節提到過，聲音的音量是由聲波的震幅來決定，因此我們可藉由震幅的大小來改變音量，例如： playVolume01.m [y, fs]=audioread('welcome.wav'); p=audioplayer(1*y, fs); playblocking(p); % 播放 1 倍震幅的音訊 p=audioplayer(3*y, fs); playblocking(p); % 播放 3 倍震幅的音訊 p=audioplayer(5*y, fs); playblocking(p); % 播放 5 倍震幅的音訊 聲音聽起來並沒有變成3或5倍大聲，為什麼？

Change of Audio Amplitude Volume of audio signals is determine by their amplitude. Here are modify amplitude to change the volume: playVolume01.m [y, fs]=audioread('welcome.wav'); p=audioplayer(1*y, fs); playblocking(p); % Original audio p=audioplayer(3*y, fs); playblocking(p); % Audio of 3x amplitude p=audioplayer(5*y, fs); playblocking(p); % Audio of 5x amplitude The playback doesn’t sound like 5-times louder, why?

改變音訊播放的取樣率 (1/2) 如果在播放時，改變取樣頻率，就會改變整個音訊的時間長度，進而影響到音高。 在下例中，我們漸漸提高播放時的取樣頻率，聽到的聲音就會越來越快、越來越高，最後出現像唐老鴨的聲音。為什麼？ playFs01.m [y, fs]=audioread('welcome.wav'); p=audioplayer(y, fs); p.SampleRate=1.0*fs; playblocking(p); % 播放 1.0 倍速度的音訊 p.SampleRate=1.2*fs; playblocking(p); % 播放 1.2 倍速度的音訊 p.SampleRate=1.5*fs; playblocking(p); % 播放 1.5 倍速度的音訊 p.SampleRate=2.0*fs; playblocking(p); % 播放 2.0 倍速度的音訊

Change of Sample Rates (1/2) Change of sample rate during playback  Change of duration  Change of the perceived pitch Increase the sample rate during playback, and you’ll hear Donald Duck (唐老鴨). Why? playFs01.m [y, fs]=audioread('welcome.wav'); p=audioplayer(y, fs); p.SampleRate=1.0*fs; playblocking(p); % Duration ratio: 1 p.SampleRate=1.2*fs; playblocking(p); % Duration ratio: 1/1.2 p.SampleRate=1.5*fs; playblocking(p); % Duration ratio: 1/1.5 p.SampleRate=2.0*fs; playblocking(p); % Duration ratio: 1/2

改變音訊播放的取樣率 (2/2) 反之，如果漸漸降低播放的頻率，聽到的聲音就會越來越慢、越來越低，最後出現像牛叫的聲音。 playFs02.m [y, fs]=audioread('welcome.wav'); p=audioplayer(y, fs); p.SampleRate=1.0*fs; playblocking(p); % 播放 1.0 倍速度的音訊 p.SampleRate=0.9*fs; playblocking(p); % 播放 0.9 倍速度的音訊 p.SampleRate=0.8*fs; playblocking(p); % 播放 0.8 倍速度的音訊 p.SampleRate=0.6*fs; playblocking(p); % 播放 0.6 倍速度的音訊

Change of Sample Rates (2/2) On the other hand, decrease the sample rate during playback, and you’ll hear cow moo… playFs02.m [y, fs]=audioread('welcome.wav'); p=audioplayer(y, fs); p.SampleRate=1.0*fs; playblocking(p); % Duration ratio: 1 p.SampleRate=0.9*fs; playblocking(p); % Duration ratio: 1/0.9 p.SampleRate=0.8*fs; playblocking(p); % Duration ratio: 1/0.8 p.SampleRate=0.6*fs; playblocking(p); % Duration ratio: 1/0.6

Observations Observations How to… Larger sample rate for playback leads to… Shorter duration and higher pitch Smaller sample rate for playback leads to... Longer duration and lower pitch How to… Generate higher pitch without duration change?  Pitch modification Create longer duration without pitch change?  Duration modification Demo Quiz!

改變符號及改變時序 如果我們將聲波訊號上下顛倒，聽到的聲音基本上是一樣的，但是如果前後顛倒，聽到的聲音就如同錄音帶「倒放」的聲音，聽起來很像是某種外國語音，請試試下列範例： 範例20-11： playReverse01.m [y, fs]=audioread('welcome.wav'); p=audioplayer(y, fs); playblocking(p); % 播放正常的音訊波形 p=audioplayer(-y, fs); playblocking(p); % 播放上下顛倒的音訊波形 p=audioplayer(flipud(y), fs); playblocking(p); % 播放前後顛倒的音訊波形

Change of Sign & Time Sequence Change of sign  No perceptional difference Reverse sequence  Sounds like another spoken language? 範例20-11： playReverse01.m [y, fs]=audioread('welcome.wav'); p=audioplayer(y, fs); playblocking(p); % Normal playback p=audioplayer(-y, fs); playblocking(p); % Change of sign p=audioplayer(flipud(y), fs); playblocking(p); % Reverse the sequence

同步及非同步播放 通常在播放音訊時，MATLAB 停止進行其他動作，直到音訊播放完畢後，才會再進行其他指令的運算，此種運作方式稱為「同步式」（Synchronous）。若需要一邊播放、一邊進行其他運算，就必須使用「非同步式」（Asynchronous）的播放方式。 範例20-12： playSync01.m [y, fs]=audioread('welcome.wav'); p=audioplayer(y, fs); playblocking(p); % 同步播放 1.0 倍速度的音訊 sound(y, 0.8*fs); % 非同步播放 0.8 倍速度的音訊 sound(y, 0.6*fs); % 非同步播放 0.6 倍速度的音訊

非同步播放 在此例中，我們會聽到類似男女兩部合唱，一快一慢，這是因為 sound 指令的預設播放方式就是「非同步」。 範例20-13： playSync02.m load handel.mat sound(y, Fs); sound(y, 1.2*Fs);

Playback Modes There are two playback modes playSync01.m playSync02.m Synchronous mode: Block everything till the end of playback. Asychronous mode: Nonblocking playSync01.m playSync02.m [y, fs]=audioread('welcome.wav'); p=audioplayer(y, fs); playblocking(p); % Synchronous sound(y, 0.8*fs); % Asynchronous sound(y, 0.6*fs); % Asynchronous load handel.mat sound(y, Fs); sound(y, 1.2*Fs);

音量自動調整 另一個類似的指令是 soundsc，可針對音訊變數的數值先進行正規化（介於 –1 和 1 中間）後，再送到喇叭播放，以達到最好的播放效果。 soundsc01.m [y, fs]=audioread('welcome.wav'); sound(y, fs); fprintf('Press any key to continue...\n'); pause soundsc(y, fs); 在影像顯示方面，對應的命令是 imagesc。

Automatic Volume Adjustment “soundsc” adjusts the volume (by normalizing the signals to have max of 1 or min of -1) before playback soundsc01.m [y, fs]=audioread('welcome.wav'); sound(y, fs); fprintf('Press any key to continue...\n'); pause soundsc(y, fs); The corresponding command for image display is “imagesc”.

聲音訊號的錄製 我們在第一節已經說明了如何讀取音訊檔案，並在第二節說明如何播放。MATLAB 也支援直接由麥克風讀取訊號，因此可以直接進行聲音的錄製，所使用的指令是 audiorecorder recordblocking

Recording of Audio Signals We can use the following MATLAB commands for recording from the microphone directly: audiorecorder recordblocking

音訊的錄製範例 (1/2) 使用預設參數，由麥克風進行3秒錄音： audioRecord01.m 預設錄音參數 取樣頻率為 8000 Hz 取樣點解析度為 8 bits 單聲道錄音 duration=3; % 錄音時間 recObj=audiorecorder; fprintf('按任意鍵後開始 %g 秒錄音：', duration); pause fprintf('錄音中...'); recordblocking(recObj, duration); fprintf('錄音結束\n'); fprintf('按任意鍵後開始播放：'); pause play(recObj);

Recording (1/2) Use default setting for 3-sec recording: audioRecord01.m Default settings for recording Sample rate: 8000 Hz Bit resolution: 8 bits Mono duration=3; % Duration of recording recObj=audiorecorder; fprintf('按任意鍵後開始 %g 秒錄音：', duration); pause % Prompt fprintf('錄音中...'); % During recording recordblocking(recObj, duration); fprintf('錄音結束\n'); % End of recording fprintf('按任意鍵後開始播放：'); pause % Press any key for playback play(recObj);

音訊的錄製範例 (2/2) 設定各項錄音參數來進行3秒錄音，並畫出波形： audioRecord02.m 設定錄音參數 取得音訊資料 fs=16000; % 取樣頻率 nBits=16; % 取樣點解析度，必須是 8 或 16 或 24 nChannel=1; % 聲道個數，必須是1(單聲道)或2(雙聲道或立體音) duration=3; % 錄音時間（秒） recObj=audiorecorder(fs, nBits, nChannel); fprintf('按任意鍵後開始 %g 秒錄音：', duration); pause fprintf('錄音中...'); recordblocking(recObj, duration); fprintf('錄音結束\n'); fprintf('按任意鍵後開始播放：'); pause play(recObj); y = getaudiodata(recObj, 'double'); % get data as a double array plot((1:length(data))/fs, y); xlabel('Time (sec)'); ylabel('Amplitude'); 設定錄音參數 取得音訊資料

Recording (2/2) Set recording parameters, record, plot the waveform: audioRecord02.m fs=16000; % Sample rate nBits=16; % Bit resolution (must be 8, 16, or 24) nChannel=1; % No. of channels (must be 1 or 2) duration=3; % Duration for recording in sec recObj=audiorecorder(fs, nBits, nChannel); fprintf('按任意鍵後開始 %g 秒錄音：', duration); pause fprintf('錄音中...'); recordblocking(recObj, duration); fprintf('錄音結束\n'); fprintf('按任意鍵後開始播放：'); pause play(recObj); y = getaudiodata(recObj, 'double'); % get data as a double array plot((1:length(data))/fs, y); xlabel('Time (sec)'); ylabel('Amplitude'); Set up recording parameters Obtain audio signals

聲音訊號的寫檔 (1/2) 我們也可以經由 MATLAB 將音訊資料直接儲存為音訊檔案，以便直接在電腦播放。寫入音訊檔案的指令是 audiowrite，其格式為: audiowrite(audioFile, y, fs) audioFile 則是欲寫入資料的檔案名稱，y 是音訊變數，fs 是取樣頻率。 範例：audioWrite01.m load handel.mat audioFile='handel.wav'; % 欲儲存的 wav 檔案 fprintf('Saving to %s...\n', audioFile); audiowrite(audioFile, y, round(1.5*Fs)); fprintf('按任意鍵後開始播放 %s...\n', audioFile); dos(['start ', audioFile]); % 開啟與 wav 檔案對應的應用程式

Storing Audio Files (1/2) We can use “audiowrite” to save audio files, with the following I/O format: audiowrite(audioFile, y, fs) audioFile: file to write to, y: audio signals, fs: sample rate 範例：audioWrite01.m load handel.mat audioFile='handel.wav'; % wav file to write to fprintf('Saving to %s...\n', audioFile); audiowrite(audioFile, y, round(1.5*Fs)); fprintf('按任意鍵後開始播放 %s...\n', audioFile); dos(['start ', audioFile]); % Use default application to open the wav file

聲音訊號的寫檔 (2/2) 錄音、播放、存檔的範例： 範例：audioWrite02.m fs=16000; % 取樣頻率 nBits=16; % 取樣點解析度，必須是 8 或 16 或 24 nChannel=1; % 聲道個數，必須是 1（單聲道） 或 2（雙聲道或立體音） duration=3; % 錄音時間（秒） recObj=audiorecorder(fs, nBits, nChannel); fprintf('按任意鍵後開始 %g 秒錄音：', duration); pause fprintf('錄音中...'); recordblocking(recObj, duration); fprintf('錄音結束\n'); fprintf('按任意鍵後開始播放：\n'); pause y = getaudiodata(recObj, 'double'); % get data as a double array plot((1:length(data))/fs, y); xlabel('Time (sec)'); ylabel('Amplitude'); sound(y, fs); audioFile='myRecording.wav'; % 欲儲存的 wav 檔案 fprintf('Saving to %s...\n', audioFile); audiowrite(audioFile, y, fs); system(audioFile); % Use default application to open the wav file

Storing Audio Files (2/2) Example of recording, playback, and saving: 範例：audioWrite02.m fs=16000; % Sample rate nBits=16; % Bit resolution (must be 8, 16, or 24) nChannel=1; % No. of channels (must be 1 or 2) duration=3; % Recording duration in sec recObj=audiorecorder(fs, nBits, nChannel); fprintf('按任意鍵後開始 %g 秒錄音：', duration); pause fprintf('錄音中...'); recordblocking(recObj, duration); fprintf('錄音結束\n'); fprintf('按任意鍵後開始播放：\n'); pause y = getaudiodata(recObj, 'double'); % get data as a double array plot((1:length(data))/fs, y); xlabel('Time (sec)'); ylabel('Amplitude'); sound(y, fs); audioFile='myRecording.wav'; % wav file to be saved fprintf('Saving to %s...\n', audioFile); audiowrite(audioFile, y, fs); system(audioFile); % 開啟與 wav 檔案對應的應用程式

Cross-version Issues File mapping for different versions of MATLAB wavread  audioread wavwrite  audiowrite wavplay  audioplayer, sound wavrecord  audiorecorder Other supports audiodevinfo playblocking play SAP toolbox Version-independent audio file reading  myAudioRead.m Progressive bar  audioPlayWithBar.m

Supplementary Material Other resources ASPR: Audio Signal Processing and Recognition Texts for this set of slides can be found at Chapter 4. Pitch tracking by visual inspection can be found at Section 4 of Chapter 5. (Slides)