Detect a specific frequency/tone from raw wave-dat

问题:

I am reading a raw wave stream coming from the microphone.
(This part works as I can send it to the speaker and get a nice echo.)

For simplicity lets say I want to detect a DTMF-tone in the wave data. In reality I want to detect any frequency, not just those in DTMF. But I always know which frequency I am looking for.

I have tried running it through FFT, but it doesn't seem very efficient if I want high accuracy in the detection (say it is there for only 20 ms). I can detect it down to an accuracy of around 200 ms.

What are my options with regards to algorithms? Are there any .Net libs for it?

回答1:

You may want to look at the Goertzel algorithm if you're trying to detect specific frequencies such as DTMF input. There is a C# DTMF generator/detector library on Sourceforge based on this algorithm.

回答2:

Very nice implementation of Goertzel is there. C# modification:

private double GoertzelFilter(float[] samples, double freq, int start, int end)
    {
        double sPrev = 0.0;
        double sPrev2 = 0.0;
        int i;
        double normalizedfreq = freq / SIGNAL_SAMPLE_RATE;
        double coeff = 2 * Math.Cos(2 * Math.PI * normalizedfreq);
        for (i = start; i < end; i++)
        {
            double s = samples[i] + coeff * sPrev - sPrev2;
            sPrev2 = sPrev;
            sPrev = s;
        }
        double power = sPrev2 * sPrev2 + sPrev * sPrev - coeff * sPrev * sPrev2;
        return power;
    }

Works great for me.

回答3:

I found this as a simple implementation of Goertzel. Haven't gotten it to work yet (looking for wrong frequency?), but I thought I'd share it anywas. It is copied from this site.

        public static double CalculateGoertzel(byte[] sample, double frequency, int samplerate)
        {
            double Skn, Skn1, Skn2;
            Skn = Skn1 = Skn2 = 0;
            for (int i = 0; i < sample.Length; i++)
            {
                Skn2 = Skn1;
                Skn1 = Skn;
                Skn = 2 * Math.Cos(2 * Math.PI * frequency / samplerate) * Skn1 - Skn2 + sample[i];
            }
            double WNk = Math.Exp(-2 * Math.PI * frequency / samplerate);
            return 20 * Math.Log10(Math.Abs((Skn - WNk * Skn1)));
        }

回答4:

Let's say that typical DTMF frequency is 200Hz - 1000Hz. Then you'd have to detect a signal based on between 4 and 20 cycles. FFT will not get you anywhere I guess, since you'll detect only multiples of 50Hz frequencies: this is a built in feature of FFT, increasing the number of samples will not solve your problem. You'll have to do something more clever.

Your best shot is to linear least-square fit your data to

h(t) = A cos (omega t) + B sin (omega t)

for a given omega (one of the DTMF frequencies). See this for details (in particular how to set a statistical significance level) and links to the litterature.

回答5:

As far as any .NET libraries that do this try TAPIEx ToneDecoder.Net Component. I use it for detecting DTMF, but it can do custom tones as well.

I know this question is old, but maybe it will save someone else several days of searching and trying out code samples and libraries that just don't work.

回答6:

Spectral Analysis.

All application where you extract frequencies from signals goes under field spectral analysis.