Swift FFT - a difficult separation problem

Question

I am trying to execute FFT in an audio file to find the frequency using the Accelerate framework. I applied the code (possibly incorrect) from this question: Spectrogram from AVAudioPCMBuffer using acceleration in Swift

Despite the fact that the values ​​from the “spectrum” are either “0”, “inf”, or “nan”, the “real” and “imaginary” components of complex separation print similar results; that this is causing the problem: 'value = sqrt (pow ( real , 2) + pow ( imag , 2)'. Correct me if i 'is wrong, but I think the rest of the code is fine.

Why do I get these results and how can I fix them (what should the separation components be) and what am I doing wrong? Please keep in mind that I am very new to FFT and fetch and have no idea how to configure this for an audio file, so any help would be greatly appreciated. Thank you

Here is the code I'm using:

// get audio file let fileURL:NSURL = NSBundle.mainBundle().URLForResource("foo", withExtension: "mp3")! let audioFile = try! AVAudioFile(forReading: fileURL) let fileFormat = audioFile.processingFormat let frameCount = UInt32(audioFile.length) let buffer = AVAudioPCMBuffer(PCMFormat: fileFormat, frameCapacity: frameCount) let audioEngine = AVAudioEngine() let playerNode = AVAudioPlayerNode() audioMixerNode = audioEngine.mainMixerNode let bufferSize = Int(frameCount) let channels: NSArray = [Int(buffer.format.channelCount)] let channelCount = channels.count let floats1 = [Int(buffer.frameLength)] for var i=0; i<channelCount; ++i { channelSamples.append([]) let firstSample = buffer.format.interleaved ? i : i*bufferSize for var j=firstSample; j<bufferSize; j+=buffer.stride*2 { channelSamples[i].append(DSPComplex(real: buffer.floatChannelData.memory[j], imag: buffer.floatChannelData.memory[j+buffer.stride])) } } // connect node audioEngine.attachNode(playerNode) audioEngine.connect(playerNode, to: audioMixerNode, format: playerNode.outputFormatForBus(0)) // Set up the transform let log2n = UInt(round(log2(Double(bufferSize)))) let fftSetup = vDSP_create_fftsetup(log2n, Int32(kFFTRadix2)) // Create the complex split value to hold the output of the transform // why doesn't this work? var realp = [Float](count: bufferSize/2, repeatedValue: 0) var imagp = [Float](count: bufferSize/2, repeatedValue: 0) var output = DSPSplitComplex(realp: &realp, imagp: &imagp) vDSP_ctoz(UnsafePointer(channelSamples), 2, &output, 1, UInt(bufferSize / 2)) // Do the fast Fourier forward transform vDSP_fft_zrip(fftSetup, &output, 1, log2n, Int32(FFT_FORWARD)) // Convert the complex output to magnitude var fft = [Float](count:Int(bufferSize / 2), repeatedValue:0.0) let bufferOver2: vDSP_Length = vDSP_Length(bufferSize / 2) vDSP_zvmags(&output, 1, &fft, 1, bufferOver2) var spectrum = [Float]() for var i=0; i<bufferSize/2; ++i { let imag = output.imagp[i] let real = output.realp[i] let magnitude = sqrt(pow(real,2)+pow(imag,2)) spectrum.append(magnitude) } // Release the setup vDSP_destroy_fftsetup(fftSetup) 

Answer 1

There were several problems with your code:

1. you did not read audio sample files
2. channelSamples was not packaged correctly
3. vDSP_fft_zrip read outside the array. he expects 2 ^ log2n samples
4. vDSP_fft_zrip packed, and your calculations are awaiting unpacking

Swift 4 version now with fix for item 3

 let fileURL = Bundle.main.url(forResource: "foo", withExtension: "mp3")! let audioFile = try! AVAudioFile(forReading: fileURL as URL) let frameCount = UInt32(audioFile.length) let log2n = UInt(round(log2(Double(frameCount)))) let bufferSizePOT = Int(1 << log2n) let buffer = AVAudioPCMBuffer(pcmFormat: audioFile.processingFormat, frameCapacity: AVAudioFrameCount(bufferSizePOT))! try! audioFile.read(into: buffer, frameCount:frameCount) // Not sure if AVAudioPCMBuffer zero initialises extra frames, so when in doubt... let leftFrames = buffer.floatChannelData![0] for i in Int(frameCount)..<Int(bufferSizePOT) { leftFrames[i] = 0 } // Set up the transform let fftSetup = vDSP_create_fftsetup(log2n, Int32(kFFTRadix2))! // create packed real input var realp = [Float](repeating: 0, count: bufferSizePOT/2) var imagp = [Float](repeating: 0, count: bufferSizePOT/2) var output = DSPSplitComplex(realp: &realp, imagp: &imagp) leftFrames.withMemoryRebound(to: DSPComplex.self, capacity: bufferSizePOT / 2) { vDSP_ctoz($0, 2, &output, 1, UInt(bufferSizePOT / 2)) } // Do the fast Fourier forward transform, packed input to packed output vDSP_fft_zrip(fftSetup, &output, 1, log2n, Int32(FFT_FORWARD)) // you can calculate magnitude squared here, with care // as the first result is wrong! read up on packed formats var fft = [Float](repeating:0.0, count:Int(bufferSizePOT / 2)) vDSP_zvmags(&output, 1, &fft, 1, vDSP_Length(bufferSizePOT / 2)) // Release the setup vDSP_destroy_fftsetup(fftSetup)