Split AKAudioFile into chunks separated by silence - ios

Question

Given a single AKAudioFile that has been created from an AKNodeRecorder containing a series of spoken words, where each word is separated by at least 1 second, what is the best approach to ultimately create a series of files with each file containing one word?
I believe this can be accomplished if there is a way to iterate the file in, for example, 100 ms chunks, and measure the average amplitude of each chunk. "Silent chunks" could be those below some arbitrarily small amplitude. While iterating, if I encounter a chunk with non-silent amplitude, I can grab the starting timestamp of this "non-silent" chunk to create an audio file that starts here and ends at the start time of the next "silent" chunk.
Whether it'd be using a manual approach like the one above or a more built-in processing technique to AudioKit, any suggestions would be greatly appreciated.

Answer 1

I don't have a complete solution, but I've started working on something similar to this. This function could serve as a jumping off point for what you need. Basically you want to read the file into a buffer then analyze the buffer data. At that point you could dice it up into smaller buffers and write those to file.
public class func guessBoundaries(url: URL, sensitivity: Double = 1) -> [Double]? {
var out: [Double] = []
guard let audioFile = try? AVAudioFile(forReading: url) else { return nil }
let processingFormat = audioFile.processingFormat
let frameCount = AVAudioFrameCount(audioFile.length)
guard let pcmBuffer = AVAudioPCMBuffer(pcmFormat: processingFormat, frameCapacity: frameCount) else { return nil }
audioFile.framePosition = 0
do {
audioFile.framePosition = 0
try audioFile.read(into: pcmBuffer, frameCount: frameCount)
} catch let err as NSError {
AKLog("ERROR: Couldn't read data into buffer. \(err)")
return nil
}
let channelCount = Int(pcmBuffer.format.channelCount)
let bufferLength = 1024
let inThreshold: Double = 0.001 / sensitivity
let outThreshold: Double = 0.0001 * sensitivity
let minSegmentDuration: Double = 1
var counter = 0
var thresholdCrossed = false
var rmsBuffer = [Float](repeating: 0, count: bufferLength)
var lastTime: Double = 0
AKLog("inThreshold", inThreshold, "outThreshold", outThreshold)
for i in 0 ..< Int(pcmBuffer.frameLength) {
// n is the channel
for n in 0 ..< channelCount {
guard let sample: Float = pcmBuffer.floatChannelData?[n][i] else { continue }
if counter == rmsBuffer.count {
let time: Double = Double(i) / processingFormat.sampleRate
let avg = rmsBuffer.reduce(0, +) / rmsBuffer.count
// AKLog("Average Value at frame \(i):", avg)
if avg > inThreshold && !thresholdCrossed && time - lastTime > minSegmentDuration {
thresholdCrossed = true
out.append(time)
lastTime = time
} else if avg <= outThreshold && thresholdCrossed && time - lastTime > minSegmentDuration {
thresholdCrossed = false
out.append(time)
lastTime = time
}
counter = 0
}
rmsBuffer[counter] = abs(sample)
counter += 1
}
}
rmsBuffer.removeAll()
return out
}