TarsosDSP 音高偏移伪像

如何解决TarsosDSP 音高偏移伪像

我正在尝试读取 .wav 文件，改变音高，然后将其写入 android 上的新文件。我正在使用 tarsosdsp 来执行此操作，但在新文件中不断出现奇怪的工件。我尝试了单个正弦波，这是文件的前几个样本的比较方式：

橙色为原始文件，蓝色由tarsos生成。据我所知，与 github 上的音高移位示例相比，我没有处理差异。这是我的代码：

    File directory = getDir("audio",Context.MODE_PRIVATE);

    String source = directory.toString() + "/" + "in.wav";
    String target = directory.toString() + "/" + "out.wav";
    RandomAccessFile outputFile = null;
    try {
        outputFile = new RandomAccessFile(target,"rw");
    } catch (FileNotFoundException e) {
        e.printstacktrace();
    }

    int bitsPerSample = 16;
    int channels = 1;
    tarsosdspAudioFormat format = new tarsosdspAudioFormat(SAMPLE_RATE,bitsPerSample,channels,true,false);

    double cents = 0;
    double factor = centToFactor(cents);
    RateTransposer rateTransposer = new RateTransposer(factor);
    WaveformSimilarityBasedOverlapAdd wsola = new WaveformSimilarityBasedOverlapAdd(WaveformSimilarityBasedOverlapAdd.Parameters.musicDefaults(factor,SAMPLE_RATE));

    WriterProcessor writer = new WriterProcessor(format,outputFile);

    try {
        InputStream inputStream = new FileInputStream(source);
        UniversalAudioInputStream uStream = new UniversalAudioInputStream(inputStream,format);
        print(wsola.getInputBufferSize());
        print(wsola.getoverlap());

        Audiodispatcher audiodispatcher = new Audiodispatcher(uStream,wsola.getInputBufferSize(),wsola.getoverlap());
        wsola.setdispatcher(audiodispatcher);
        audiodispatcher.addAudioProcessor(wsola);
        audiodispatcher.addAudioProcessor(rateTransposer);
        audiodispatcher.addAudioProcessor(writer);
        audiodispatcher.run();
    } catch (FileNotFoundException e) {
        e.printstacktrace();
    }

和：

public static double centToFactor(double cents) {
    return 1 / Math.pow(Math.E,cents * Math.log(2) / 1200 / Math.log(Math.E));
}

我尝试对声音 (cents != 0) 进行实际更改，但结果并没有好转。我是否必须希望这些伪像在真实录音中总是不明显，或者有没有办法解决这个问题？我对音频处理很陌生。