Question

我正在开发一个分析音频文件频率的简单程序。使用fft长度为8192，采样率为44100，如果我使用恒定频率的wav文件作为输入 - 比如65Hz，200Hz或300Hz - 输出是该值的常数图。

如果我使用某人说话的录音，频率的峰值高达4000Hz，平均值为450 + ish的90秒文件。起初我以为这是因为录音是立体声，但是将它转换为单声道，其比特率与测试文件完全相同，并没有太大变化。（平均值从492下降到456但仍然太高）

有没有人知道可能导致这种情况的原因？

我认为我不应该找到最高价值但可能采取平均值或中值？

编辑：使用每8192字节缓冲区的大小的平均值，并获得最接近该大小的索引会使所有内容混乱。

这是Sample Aggregator在为当前缓冲区计算fft时触发的事件处理程序的代码

void FftCalculated(object sender, FftEventArgs e)
    {
        int length = e.Result.Length;
        float[] magnitudes = new float[length];

        for (int i = 0; i < length / 2; i++)
        {
            float real = e.Result[i].X;
            float imaginary = e.Result[i].Y;
            magnitudes[i] = (float)(10 * Math.Log10(Math.Sqrt((real * real) + (imaginary * imaginary))));
        }

        float max_mag = float.MinValue;
        float max_index = -1;
        for (int i = 0; i < length / 2; i++)
            if (magnitudes[i] > max_mag)
            {
                max_mag = magnitudes[i];
                max_index = i;
            }

        var currentFrequency = max_index * SAMPLERATE / 8192; 

        Console.WriteLine("frequency be " + currentFrequency);
    }

附加：这是读取文件并将其发送到分析部分的代码

using (var rdr = new WaveFileReader(audioFilePath))
            {
                var newFormat = new WaveFormat(Convert.ToInt32(SAMPLERATE/*44100*/), 16, 1);
                byte[] buffer = new byte[8192];
                var audioData = new AudioData(); //custom class for project
                using (var conversionStream = new WaveFormatConversionStream(newFormat, rdr))
                {
                    // Used to send audio in realtime, it's a timestamps issue for the graphs 
                    // I'm working on fixing this, but it has lower priority so disregard it :p
                    TimeSpan audioDuration = conversionStream.TotalTime;
                    long audioLength = conversionStream.Length;
                    int waitTime = (int)(audioDuration.TotalMilliseconds / audioLength * 8192);
                    while (conversionStream.Read(buffer, 0, buffer.Length) != 0)
                    {
                        audioData.AudioDataBase64 = Utils.Base64Encode(buffer);
                        Thread.Sleep(waitTime);
                        SendMessage("AudioData", Utils.StringToAscii(AudioData.GetJSON(audioData)));
                    }
                    Console.WriteLine("Reached End of File");
                }
            }

这是接收音频数据的代码

{
    var audioData = new AudioData();
    audioData = 
    AudioData.GetStateFromJSON(Utils.AsciiToString(receivedMessage));
    QueueAudio(Utils.Base64Decode(audioData.AudioDataBase64)));
}

接着是

var waveFormat = new WaveFormat(Convert.ToInt32(SAMPLERATE/*44100*/), 16, 1);
_bufferedWaveProvider = new BufferedWaveProvider(waveFormat);
_bufferedWaveProvider.BufferDuration = new TimeSpan(0, 2, 0);


{
    void QueueAudio(byte[] data)
    {
        _bufferedWaveProvider.AddSamples(data, 0, data.Length);
        if (_bufferedWaveProvider.BufferedBytes >= fftLength)
        {
            byte[] buffer = new byte[_bufferedWaveProvider.BufferedBytes];
            _bufferedWaveProvider.Read(buffer, 0, _bufferedWaveProvider.BufferedBytes);
            for (int index = 0; index < buffer.Length; index += 2)
            {
                short sample = (short)((buffer[index] | buffer[index + 1] << 8));
                float sample32 = (sample) / 32767f;
                sampleAggregator.Add(sample32);
            }
        }
    }
}

然后SampleAggregator在完成fft时触发上面的事件。

NAudio频率分析仪给出了不一致的结果

0 个答案: