Improve performance of musical note synthesis using AudioTrack

I am trying to build an android app which synthesizes and plays musical notes. I am using AudioTrack. It seems that my program is unable to fill the track buffer fast enough due to the latency in generating the audio information. I am generating the audio info from the fourier coefficients of the note. Hence before each write, the program has to create and sample 2048 sine waves each with a sample size of 2400 and sample rate of 44100!

So instead of a continuous sound I hear intermittent beeps. The Logcat gives the following warning in between beeps:

W/AudioTrack﹕ obtainBuffer() track 0x177ff80 disabled, restarting

My code is given below. Can anyone identify ways in which I could optimize the code?

package com.alantaar.alantaar;

import android.media.AudioFormat;
import android.media.AudioManager;
import android.media.AudioTrack;
import android.support.v7.app.ActionBarActivity;
import android.util.Log;

import java.util.Arrays;

public class AudioGenerator {
    private AudioTrack audio;
    private double[] coefficients;
    private int bufferSize;
    private int sampleRate;
    private double[] phases;
    private double frequency;


public AudioGenerator(double[] coefficients, double frequency, int sampleRate, int bufferSize){
    int minBufferSize = AudioTrack.getMinBufferSize(sampleRate, AudioFormat.CHANNEL_OUT_MONO,
            AudioFormat.ENCODING_PCM_16BIT);
    this.bufferSize = bufferSize > minBufferSize ? bufferSize : minBufferSize;
    audio = new AudioTrack(AudioManager.STREAM_MUSIC, sampleRate, AudioFormat.CHANNEL_OUT_MONO,
            AudioFormat.ENCODING_PCM_16BIT, this.bufferSize, AudioTrack.MODE_STREAM);

    this.coefficients = coefficients;
    this.sampleRate = sampleRate;
    this.frequency = frequency;
    this.phases = new double[coefficients.length];
    Arrays.fill(this.phases, 0.0);
    Log.i("length", Integer.toString(coefficients.length));

    audio.play();
}

public void playNote(){
    while(true){
        short[] waveSample = sampleWave();
        audio.write(waveSample, 0, waveSample.length);
    }
}

private short[] sampleWave(){
    short [] waveSample = new short[bufferSize];
    Arrays.fill(waveSample, (short)0);
    for(int i = 0; i < coefficients.length; i++){
        double coefficient = coefficients[i];
        short[] sineSamples = sampleSineWave(coefficient, i);
        for(int j = 0; j < waveSample.length; j++){
            waveSample[j] += sineSamples[j];
        }
    }
    return waveSample;
}

private short [] sampleSineWave(double coefficient, int index){
    double freq = frequency * index;
    short [] samples = new short[bufferSize];
    for(int i = 0; i < bufferSize; i++){
        samples[i] = (short) (coefficient * Short.MAX_VALUE * Math.cos(phases[index]));
        phases[index] += 2 * Math.PI * freq/sampleRate;
    }
    return samples;
}

public void pauseNote(){
    audio.stop();
}

public void stopNote(){
    audio.release();
}

}

---

Here are a few things that pop out at me:

Move computations out of the inner loop. For example, the 2*pi*freq/sampleRate could be done right before the for . Same with coefficient * Short.MAX_VALUE .

I'd bet that your note doesn't actually contain 2048 different meaningful sine components.One option would be to NOT generate tones when the coefficient falls below a certain threshold.

Use FFT! You are implementing a DFT which is O(n^2). FFT runs in O(n log n).

Use the time while audio.write is blocking to generate more samples. I don't know the behavior of AudioTrack 's streaming mode but there is a good chance that your program is spending most of its time in a blocking call.

链接地址: http://www.djcxy.com/p/33862.html

上一篇: 保存Android股票语音识别引擎的音频输入

下一篇: 使用AudioTrack提高音符合成的性能