package audio import ( "math" "math/cmplx" "github.com/BeatGlow/beatglow/internal/calc" "gonum.org/v1/gonum/dsp/fourier" ) // FrequencySpectrumAnalyzer structure to hold analysis parameters type FrequencySpectrumAnalyzer struct { SampleRate float64 BitDepth int NumBins int MinFrequency float64 MaxFrequency float64 avgSamples []float64 coefficients []complex128 bins []float64 } // dBFS calculates dBFS for the given amplitude value func dBFS(amplitude float64) float64 { if amplitude == 0 { return -math.Inf(1) // -Infinity for silence } return 20 * math.Log10(amplitude) } func normalize(values []float64) { var max float64 for _, v := range values { if v = math.Abs(v); v > max { max = v } } if max == 0 { return } for i := range values { values[i] /= max } } // AnalyzeFrequencySpectrum performs the FFT and returns the energy levels in dBFS for each frequency bin func (a *FrequencySpectrumAnalyzer) AnalyzeFrequencySpectrum(buffer *Buffer) []float64 { samples := len(buffer.Samples[0]) //if len(a.avgSamples) != samples { a.avgSamples = make([]float64, samples) a.coefficients = make([]complex128, samples/2+1) //} channels := len(buffer.Samples) if channels == 1 { for i, sample := range buffer.Samples[0] { a.avgSamples[i] = float64(sample) } } else { for i := range buffer.Samples[0] { a.avgSamples[i] = float64(buffer.Samples[0][i]+buffer.Samples[1][i]) / 2 } } // Apply windowing function //normalize(a.avgSamples) //calc.Hamming(a.avgSamples, a.avgSamples) calc.Hann(a.avgSamples, a.avgSamples) // Perform FFT on the signal fft := fourier.NewFFT(samples) fft.Coefficients(a.coefficients, a.avgSamples) // Calculate the energy levels in dBFS for each frequency bin var ( frequencyBinSize = a.SampleRate / float64(samples) minBin = int(math.Round(a.MinFrequency / frequencyBinSize)) maxBin = int(math.Round(a.MaxFrequency / frequencyBinSize)) ) // Make sure the minimum bin is not lower than 0, and the maximum bin doesn't exceed half of the buffer size if minBin < 0 { minBin = 0 } if maxBin > samples/2 { maxBin = samples / 2 } //if len(a.bins) != a.NumBins { a.bins = make([]float64, a.NumBins) //} // Compute dBFS levels for the selected frequency range for i := 0; i < a.NumBins; i++ { var ( // Calculate the frequency range for this bin frequencyScale = (a.MaxFrequency - a.MinFrequency) / float64(a.NumBins) frequencyStart = a.MinFrequency + float64(i)*frequencyScale frequencyEnd = a.MinFrequency + float64(i+1)*frequencyScale // Determine the FFT bin range corresponding to the frequency range of this bin startBin = int((frequencyStart / frequencyBinSize) + .5) endBin = int((frequencyEnd / frequencyBinSize) + .5) ) // Ensure the bins are within the valid range if startBin < 0 { startBin = 0 } if endBin > samples/2 { endBin = samples / 2 } // Average the magnitudes of the FFT bins in this range and accumulate the result var totalAmplitude float64 var count int for bin := startBin; bin < endBin; bin++ { x := a.coefficients[bin] //amplitude := real(x)*real(x) + imag(x)*imag(x) // math.Hypot(real(x), imag(x)) amplitude := math.Min(cmplx.Abs(x), 1.0) totalAmplitude += amplitude count++ } // Calculate the average dBFS for this frequency bin if count > 0 { a.bins[i] = math.Min((totalAmplitude/float64(count))*10, 1.0) } } return a.bins }