[Swift] vDSP를 통해 사운드 밸런싱

70037 단어 음성 신호 처리tech

개시하다


vDSP는 수치 계산과 디지털 신호 처리를 신속하게 하는 프로그램 라이브러리이다.엑스코드와 함께 포장되어 있어 바로 사용 가능합니다.
그러면 vDSP에는 이중 필터vDSP.Biquad가 포함됩니다.iOS 14부터 사용할 수 있으니 시도해 보세요.

설치 예


다음은 설치 예입니다.읽기/tmp/input.wav 후 로우 패스 필터(주파수 220Hz/Q 값 1.23)를 적용한 결과/tmp/output.wav로 쓰기 처리됩니다.
import Accelerate
import AVFoundation

class FilterParameter {
    let b0: Double
    let b1: Double
    let b2: Double
    let a1: Double
    let a2: Double

    init(_ b0: Double, _ b1: Double, _ b2: Double, _ a1: Double, _ a2: Double) {
        self.b0 = b0
        self.b1 = b1
        self.b2 = b2
        self.a1 = a1
        self.a2 = a2
    }
}

class LowPassFilterParameter: FilterParameter {
    init(sampleRate: Double, frequency: Double, q: Double) {
        let w0: Double = 2.0 * Double.pi * frequency / sampleRate
        let alpha: Double = sin(w0) / (2.0 * q)

        let a0: Double = 1.0 + alpha
        let a1: Double = -2.0 * cos(w0)
        let a2: Double = 1.0 - alpha
        let b0: Double = (1.0 - cos(w0)) / 2.0
        let b1: Double = 1.0 - cos(w0)
        let b2: Double = (1.0 - cos(w0)) / 2.0

        super.init(b0 / a0, b1 / a0, b2 / a0, a1 / a0, a2 / a0)
    }
}

class HighPassFilterParameter: FilterParameter {
    init(sampleRate: Double, frequency: Double, q: Double) {
        let w0: Double = 2.0 * Double.pi * frequency / sampleRate
        let alpha: Double = sin(w0) / (2.0 * q)

        let a0: Double = 1.0 + alpha
        let a1: Double = -2.0 * cos(w0)
        let a2: Double = 1.0 - alpha
        let b0: Double = (1.0 + cos(w0)) / 2.0
        let b1: Double = -1.0 * (1.0 + cos(w0))
        let b2: Double = (1.0 + cos(w0)) / 2.0

        super.init(b0 / a0, b1 / a0, b2 / a0, a1 / a0, a2 / a0)
    }
}

class AllPassFilterParameter: FilterParameter {
    init(sampleRate: Double, frequency: Double, q: Double) {
        let w0: Double = 2.0 * Double.pi * frequency / sampleRate
        let alpha: Double = sin(w0) / (2.0 * q)

        let a0: Double = 1.0 + alpha
        let a1: Double = -2.0 * cos(w0)
        let a2: Double = 1.0 - alpha
        let b0: Double = 1.0 - alpha
        let b1: Double = -2.0 * cos(w0)
        let b2: Double = 1.0 + alpha

        super.init(b0 / a0, b1 / a0, b2 / a0, a1 / a0, a2 / a0)
    }
}

class BandPassFilterParameter: FilterParameter {
    init(sampleRate: Double, frequency: Double, width: Double) {
        let w0: Double = 2.0 * Double.pi * frequency / sampleRate
        let alpha: Double = sin(w0) * sinh(log(2.0) / 2.0 * width * w0 / sin(w0))

        let a0: Double = 1.0 + alpha
        let a1: Double = -2.0 * cos(w0)
        let a2: Double = 1.0 - alpha
        let b0: Double = alpha
        let b1: Double = 0.0
        let b2: Double = -1.0 * alpha

        super.init(b0 / a0, b1 / a0, b2 / a0, a1 / a0, a2 / a0)
    }
}

class BandRejectFilterParameter: FilterParameter {
    init(sampleRate: Double, frequency: Double, width: Double) {
        let w0: Double = 2.0 * Double.pi * frequency / sampleRate
        let alpha: Double = sin(w0) * sinh(log(2.0)/2.0 * width * w0 / sin(w0))

        let a0: Double = 1.0 + alpha
        let a1: Double = -2.0 * cos(w0)
        let a2: Double = 1.0 - alpha
        let b0: Double = 1.0
        let b1: Double = -2.0 * cos(w0)
        let b2: Double = 1.0

        super.init(b0 / a0, b1 / a0, b2 / a0, a1 / a0, a2 / a0)
    }
}

class LowShelfFilterParameter: FilterParameter {
    init(sampleRate: Double, frequency: Double, q: Double, gain: Double) {
        let w0: Double = 2.0 * Double.pi * frequency / sampleRate
        let a: Double = pow(10.0, (gain / 40.0))
        let beta: Double = a.squareRoot() / q

        let a0: Double = (a + 1.0) + (a - 1.0)*cos(w0) + beta * sin(w0)
        let a1: Double = -2.0 * ((a - 1.0) + (a + 1.0)*cos(w0))
        let a2: Double = (a + 1.0) + (a - 1.0)*cos(w0) - beta * sin(w0)
        let b0: Double = a * ((a + 1.0) - (a - 1.0)*cos(w0) + beta * sin(w0))
        let b1: Double = 2.0 * a * ((a - 1.0) - (a + 1.0)*cos(w0))
        let b2:Double =  a * ((a + 1.0) - (a - 1.0)*cos(w0) - beta * sin(w0))

        super.init(b0 / a0, b1 / a0, b2 / a0, a1 / a0, a2 / a0)
    }
}

class HighShelfFilterParameter: FilterParameter {
    init(sampleRate: Double, frequency: Double, q: Double, gain: Double) {
        let w0: Double = 2.0 * Double.pi * frequency / sampleRate
        let a: Double = pow(10.0, (gain / 40.0))
        let beta: Double = a.squareRoot() / q

        let a0: Double = (a + 1.0) - (a - 1.0)*cos(w0) + beta * sin(w0)
        let a1: Double = 2.0 * ((a - 1.0) - (a + 1.0)*cos(w0))
        let a2: Double = (a + 1.0) - (a - 1.0)*cos(w0) - beta * sin(w0)
        let b0: Double = a * ((a + 1.0) + (a - 1.0)*cos(w0) + beta * sin(w0))
        let b1: Double = -2.0 * a * ((a - 1.0) + (a + 1.0)*cos(w0))
        let b2: Double = a * ((a + 1.0) + (a - 1.0)*cos(w0) - beta * sin(w0))

        super.init(b0 / a0, b1 / a0, b2 / a0, a1 / a0, a2 / a0)
    }
}

class PeakingFilterParameter: FilterParameter {
    init(sampleRate: Double, frequency: Double, width: Double, gain: Double) {
        let w0: Double = 2.0 * Double.pi * frequency / sampleRate
        let alpha: Double = sin(w0) * sinh(log(2.0)/2.0 * width * w0 / sin(w0))
        let a: Double = pow(10.0, (gain / 40.0))

        let a0: Double = 1.0 + alpha / a
        let a1: Double = -2.0 * cos(w0)
        let a2: Double = 1.0 - alpha / a
        let b0: Double = 1.0 + alpha * a
        let b1: Double =  -2.0 * cos(w0)
        let b2: Double = 1.0 - alpha * a

        super.init(b0 / a0, b1 / a0, b2 / a0, a1 / a0, a2 / a0)
    }
}

func readPCMBuffer(url: URL) -> AVAudioPCMBuffer? {
    guard let input = try? AVAudioFile(forReading: url, commonFormat: .pcmFormatInt16, interleaved: false) else {
        return nil
    }
    guard let buffer = AVAudioPCMBuffer(pcmFormat: input.processingFormat, frameCapacity: AVAudioFrameCount(input.length)) else {
        return nil
    }
    do {
        try input.read(into: buffer)
    } catch {
        return nil
    }

    return buffer
}

func writePCMBuffer(url: URL, buffer: AVAudioPCMBuffer) throws {
    let settings: [String: Any] = [
        AVFormatIDKey: buffer.format.settings[AVFormatIDKey] ?? kAudioFormatLinearPCM,
        AVNumberOfChannelsKey: buffer.format.settings[AVNumberOfChannelsKey] ?? 2,
        AVSampleRateKey: buffer.format.settings[AVSampleRateKey] ?? 44100,
        AVLinearPCMBitDepthKey: buffer.format.settings[AVLinearPCMBitDepthKey] ?? 16
    ]

    do {
        let output = try AVAudioFile(forWriting: url, settings: settings, commonFormat: .pcmFormatInt16, interleaved: false)
        try output.write(from: buffer)
    } catch {
        throw error
    }
}

func applyLowPassFilter(from inputPath: String, to outputPath: String) {
    guard let inputBuffer = readPCMBuffer(url: URL(string: inputPath)!) else {
        fatalError("failed to read \(inputPath)")
    }
    guard let outputBuffer = AVAudioPCMBuffer(pcmFormat: inputBuffer.format, frameCapacity: inputBuffer.frameLength) else {
        fatalError("failed to create a buffer for writing")
    }
    guard let inputInt16ChannelData = inputBuffer.int16ChannelData else {
        fatalError("failed to obtain underlying input buffer")
    }
    guard let outputInt16ChannelData = outputBuffer.int16ChannelData else {
        fatalError("failed to obtain underlying output buffer")
    }

    // Calculate a filter coefficients. (Type = low-pass / Cut off frequency = 440 Hz / Q-value = 1.23)
    let parameter = LowPassFilterParameter(sampleRate: inputBuffer.format.sampleRate, frequency: 220.0, q: 1.23)

    // Because the input might be stereo, create filters for each channels.
    var filters = [vDSP.Biquad](repeating: vDSP.Biquad(coefficients: [parameter.b0, parameter.b1, parameter.b2, parameter.a1, parameter.a2], channelCount: 1, sectionCount: 1, ofType: Float.self)!, count: Int(inputBuffer.format.channelCount))

    for channel in 0 ..< Int(inputBuffer.format.channelCount) {
        let p1: UnsafeMutablePointer<Int16> = inputInt16ChannelData[channel]
        let p2: UnsafeMutablePointer<Int16> = outputInt16ChannelData[channel]

        var signal = [Float](repeating: 0.0, count: Int(inputBuffer.frameLength))

        for i in 0 ..< Int(inputBuffer.frameLength) {
            signal[i] = Float(p1[i]) / 32767.0
        }

        signal = filters[channel].apply(input: signal)

        for i in 0 ..< Int(inputBuffer.frameLength) {
            if signal[i] < -1.0 {
                p2[i] = -32767
            } else if signal[i] > 1.0 {
                p2[i] = 32767
            } else {
                p2[i] = Int16(Int(signal[i] * 32767))
            }
        }
    }

    outputBuffer.frameLength = inputBuffer.frameLength

    do {
        try writePCMBuffer(url: URL(string: outputPath)!, buffer: outputBuffer)
    } catch {
        fatalError("failed to write \(outputPath)")
    }
}

applyLowPassFilter(from: "file:///tmp/input.wav", to: "file:///tmp/output.wav")
상기 내용을 main.swift로 저장하고 다음과 같이 집행한다.
$ swift main.swift
실시례에서 저통 필터를 포함한 총 8종의 필터를 정의했다.파라미터를 가지고 놀아 주세요.

설치 고려 사항


아직 주의해야 할 정도는 아니지만 vDSP.비쿼드에게 주는 계수는 6개가 아닌 5개다.a0 정규화1.

참고 문헌

  • Equalizing Audio with vDSP - Apple Developer
  • vDSP.Biquad - Apple Developer
  • Cookbook formulae for audio EQ biquad filter coefficients
    • Regcode 소개 - 코드로 쉽게 정규식 만들기
    Java를 사용하여 Excel 데이터 정렬

    좋은 웹페이지 즐겨찾기

    개발자 우수 사이트 수집

    개발자가 알아야 할 필수 사이트 100선 추천 우리는 당신을 위해 100개의 자주 사용하는 개발자 학습 사이트를 정리했습니다
    best100-homepage