/// <summary> /// Create a new sound by changing the speed/pitch of the old sound. 2.0 = an octave up, 1.0 = the same, 0.5 = down an octave /// </summary> public static WaveAudio ScalePitchAndDuration(WaveAudio w, double factor) { if (factor < 0) throw new ArgumentException("Factor must >= 0"); WaveAudio res = new WaveAudio(w.getSampleRate(), w.getNumChannels()); // do operation for all channels for (int i = 0; i < w.getNumChannels(); i++) res.data[i] = scalePitchAndDurationChannel(w.data[i], factor); return res; }
public static WaveAudio hiPassFilter(WaveAudio w, double factor) //0.5 { WaveAudio ret = new WaveAudio(w.getSampleRate(), w.getNumChannels()); ret.LengthInSamples = w.LengthInSamples; for (int ch = 0; ch < w.getNumChannels(); ch++) { ret.data[ch][0] = w.data[ch][0]; for (int i = 1; i < ret.data[ch].Length; i++) ret.data[ch][i] = factor * ret.data[ch][i - 1] + (factor) * (w.data[ch][i] - w.data[ch][i-1]); } return ret; }
//could also get continuous with window. public static WaveAudio lowPassFilter(WaveAudio w, double factor) //0.5 { //http://en.wikipedia.org/wiki/Low-pass_filter WaveAudio ret = new WaveAudio(w.getSampleRate(), w.getNumChannels()); ret.LengthInSamples = w.LengthInSamples; for (int ch = 0; ch < w.getNumChannels(); ch++) { ret.data[ch][0] = w.data[ch][0]; for (int i=1; i<ret.data[ch].Length; i++) ret.data[ch][i] = (1-factor)*ret.data[ch][i-1] + (factor)*w.data[ch][i]; } return ret; }
public static WaveAudio hiPassFilter(WaveAudio w, double factor) //0.5 { WaveAudio ret = new WaveAudio(w.getSampleRate(), w.getNumChannels()); ret.LengthInSamples = w.LengthInSamples; for (int ch = 0; ch < w.getNumChannels(); ch++) { ret.data[ch][0] = w.data[ch][0]; for (int i = 1; i < ret.data[ch].Length; i++) { ret.data[ch][i] = factor * ret.data[ch][i - 1] + (factor) * (w.data[ch][i] - w.data[ch][i - 1]); } } return(ret); }
public static WaveAudio Tremolo(WaveAudio w, double tremfreq, double amp) { WaveAudio res = new WaveAudio(w.getSampleRate(), w.getNumChannels()); res.LengthInSamples = w.LengthInSamples; double tremeloFreqScale = 2.0 * Math.PI * tremfreq / (double)w.getSampleRate(); for (int ch = 0; ch < w.data.Length; ch++) { for (int i = 0; i < w.data[ch].Length; i++) { double val = w.data[ch][i] * (1 + amp * Math.Sin(tremeloFreqScale * i)); if (val > 1.0) { val = 1.0; } else if (val < -1.0) { val = -1.0; } res.data[ch][i] = val; } } return(res); }
public static WaveAudio Vibrato(WaveAudio wave, double freq, double width) { if (width < 0) { throw new ArgumentException("Factor must >= 0"); } WaveAudio newwave = new WaveAudio(wave.getSampleRate(), wave.getNumChannels()); // do operation for all channels for (int i = 0; i < wave.getNumChannels(); i++) { newwave.data[i] = vibratoChannel(wave.data[i], wave.getSampleRate(), width, freq); } return(newwave); }
/// <summary> /// Create a new sound by changing the speed/pitch of the old sound. 2.0 = an octave up, 1.0 = the same, 0.5 = down an octave /// </summary> public static WaveAudio ScalePitchAndDuration(WaveAudio w, double factor) { if (factor < 0) { throw new ArgumentException("Factor must >= 0"); } WaveAudio res = new WaveAudio(w.getSampleRate(), w.getNumChannels()); // do operation for all channels for (int i = 0; i < w.getNumChannels(); i++) { res.data[i] = scalePitchAndDurationChannel(w.data[i], factor); } return(res); }
//could also get continuous with window. public static WaveAudio lowPassFilter(WaveAudio w, double factor) //0.5 { //http://en.wikipedia.org/wiki/Low-pass_filter WaveAudio ret = new WaveAudio(w.getSampleRate(), w.getNumChannels()); ret.LengthInSamples = w.LengthInSamples; for (int ch = 0; ch < w.getNumChannels(); ch++) { ret.data[ch][0] = w.data[ch][0]; for (int i = 1; i < ret.data[ch].Length; i++) { ret.data[ch][i] = (1 - factor) * ret.data[ch][i - 1] + (factor) * w.data[ch][i]; } } return(ret); }
public static double[][] SpectrumContentOverTime(WaveAudio w, int nBins, int nSize) { if (w.getNumChannels() != 1) { throw new Exception("Only mono supported."); } if (!isPowerOfTwo((uint)nSize)) { throw new Exception("Size must be power of 2."); } int nDatapoints = w.LengthInSamples / nSize - 1; double[][] res = new double[nDatapoints][]; double[] buffer = new double[nSize]; for (int i = 0; i < nDatapoints; i++) { // get samples from this slice of time. Put it into the buffer Array.Copy(w.data[0], i * nSize, buffer, 0, nSize); res[i] = getSpectrumContent(buffer, nBins); } return(res); }
// The following create new audio without modifying original public static WaveAudio Concatenate(WaveAudio w1, WaveAudio w2) { // make sure sample rates match we could be nicer and convert automatically if (w1.m_currentSampleRate != w2.m_currentSampleRate) throw new Exception("Sample rates don't match"); if (w2.getNumChannels() != w2.getNumChannels()) throw new Exception("Number of channels don't match"); WaveAudio newwave = new WaveAudio(w1.getSampleRate(), w1.getNumChannels()); newwave.LengthInSamples = w1.LengthInSamples + w2.LengthInSamples; for (int ch = 0; ch < w1.getNumChannels(); ch++) { // source sIndex, destination, destIndex, length Array.Copy(w1.data[ch], 0, newwave.data[ch], 0, w1.data[ch].Length); Array.Copy(w2.data[ch], 0, newwave.data[ch], w1.data[ch].Length, w2.data[0].Length); } return newwave; }
public static WaveAudio GetSliceSample(WaveAudio wthis, int nStart, int nEnd) { WaveAudio slice = new WaveAudio(wthis.getSampleRate(), wthis.getNumChannels()); if (nEnd <= nStart || nEnd > wthis.LengthInSamples || nStart < 0) throw new Exception("Invalid slice"); for (int ch = 0; ch < slice.data.Length; ch++) { slice.data[ch] = new double[nEnd - nStart]; Array.Copy(wthis.data[ch], nStart, slice.data[ch], 0, nEnd - nStart); } return slice; }
// These work by shifting the signal until it seems to correlate with itself. // In other words if the signal looks very similar to (signal shifted 200 samples) than the fundamental period is probably 200 samples // Note that the algorithm only works well when there's only one prominent fundamental. // This could be optimized by looking at the rate of change to determine a maximum without testing all periods. private static double[] detectPitchCalculation(WaveAudio w, double minHz, double maxHz, int nCandidates, int nResolution, PitchDetectAlgorithm algorithm) { // note that higher frequency means lower period int nLowPeriodInSamples = hzToPeriodInSamples(maxHz, w.getSampleRate()); int nHiPeriodInSamples = hzToPeriodInSamples(minHz, w.getSampleRate()); if (nHiPeriodInSamples <= nLowPeriodInSamples) throw new Exception("Bad range for pitch detection."); if (w.getNumChannels() != 1) throw new Exception("Only mono supported."); double[] samples = w.data[0]; if (samples.Length < nHiPeriodInSamples) throw new Exception("Not enough samples."); // both algorithms work in a similar way // they yield an array of data, and then we find the index at which the value is highest. double[] results = new double[nHiPeriodInSamples - nLowPeriodInSamples]; if (algorithm == PitchDetectAlgorithm.Amdf) { for (int period = nLowPeriodInSamples; period < nHiPeriodInSamples; period += nResolution) { double sum = 0; // for each sample, see how close it is to a sample n away. Then sum these. for (int i = 0; i < samples.Length - period; i++) sum += Math.Abs(samples[i] - samples[i + period]); double mean = sum / (double)samples.Length; mean *= -1; //somewhat of a hack. We are trying to find the minimum value, but our findBestCandidates finds the max. value. results[period - nLowPeriodInSamples] = mean; } } else if (algorithm == PitchDetectAlgorithm.Autocorrelation) { for (int period = nLowPeriodInSamples; period < nHiPeriodInSamples; period += nResolution) { double sum = 0; // for each sample, find correlation. (If they are far apart, small) for (int i = 0; i < samples.Length - period; i++) sum += samples[i] * samples[i + period]; double mean = sum / (double)samples.Length; results[period - nLowPeriodInSamples] = mean; } } // find the best indices int[] bestIndices = findBestCandidates(nCandidates, ref results); //note findBestCandidates modifies parameter // convert back to Hz double[] res = new double[nCandidates]; for (int i=0; i<nCandidates;i++) res[i] = periodInSamplesToHz(bestIndices[i]+nLowPeriodInSamples, w.getSampleRate()); return res; }
// The following create new audio without modifying original public static WaveAudio Concatenate(WaveAudio w1, WaveAudio w2) { // make sure sample rates match we could be nicer and convert automatically if (w1.m_currentSampleRate != w2.m_currentSampleRate) { throw new Exception("Sample rates don't match"); } if (w2.getNumChannels() != w2.getNumChannels()) { throw new Exception("Number of channels don't match"); } WaveAudio newwave = new WaveAudio(w1.getSampleRate(), w1.getNumChannels()); newwave.LengthInSamples = w1.LengthInSamples + w2.LengthInSamples; for (int ch = 0; ch < w1.getNumChannels(); ch++) { // source sIndex, destination, destIndex, length Array.Copy(w1.data[ch], 0, newwave.data[ch], 0, w1.data[ch].Length); Array.Copy(w2.data[ch], 0, newwave.data[ch], w1.data[ch].Length, w2.data[0].Length); } return(newwave); }
public static WaveAudio GetSliceSample(WaveAudio wthis, int nStart, int nEnd) { WaveAudio slice = new WaveAudio(wthis.getSampleRate(), wthis.getNumChannels()); if (nEnd <= nStart || nEnd > wthis.LengthInSamples || nStart < 0) { throw new Exception("Invalid slice"); } for (int ch = 0; ch < slice.data.Length; ch++) { slice.data[ch] = new double[nEnd - nStart]; Array.Copy(wthis.data[ch], nStart, slice.data[ch], 0, nEnd - nStart); } return(slice); }
/// <summary> /// Find spectrum content of signal. Returns array, where each element represents energy at frequencies. /// For example, SpectrumContent(w, 8) returns 8 numbers. The first in the array is the amount of energy at low frequencies, and the last is the energy at highest frequencies. /// Note that FFT uses a power of 2 samples, and so all of the signal may not be used. /// </summary> /// <param name="w">Sound</param> /// <param name="nBins">Number of bins to return.</param> public static double[] SpectrumContent(WaveAudio w, int nBins) { if (w.getNumChannels() != 1) throw new Exception("Only mono supported."); double[] buffer; // FFT uses a power of 2 samples, so we might have to truncate. if (!isPowerOfTwo((uint) w.LengthInSamples)) { int nSize = (int)findNearestPowerOfTwo((uint) w.LengthInSamples); buffer = new double[nSize]; Array.Copy(w.data[0], buffer, nSize); } else buffer = w.data[0]; return getSpectrumContent(buffer, nBins); }
public static double[][] SpectrumContentOverTime(WaveAudio w, int nBins, int nSize) { if (w.getNumChannels() != 1) throw new Exception("Only mono supported."); if (!isPowerOfTwo((uint) nSize)) throw new Exception("Size must be power of 2."); int nDatapoints = w.LengthInSamples / nSize - 1; double[][] res = new double[nDatapoints][]; double[] buffer = new double[nSize]; for (int i = 0; i < nDatapoints; i++) { // get samples from this slice of time. Put it into the buffer Array.Copy(w.data[0], i * nSize, buffer, 0, nSize); res[i] = getSpectrumContent(buffer, nBins); } return res; }
public static WaveAudio Phaser(WaveAudio wOriginal, double freq, double fb, int depth, int stages, int drywet) { double startphaseleft = 0; double startphaseright = startphaseleft + Math.PI; //note that left and right channels should start pi out of phase double freq_scaled = 2.0 * Math.PI * freq / (double)wOriginal.getSampleRate(); WaveAudio w = wOriginal.Clone(); if (w.getNumChannels() == 1) { effect_phaseraud_impl(w.data[0], freq_scaled, startphaseleft, fb, depth, stages, drywet); } else { effect_phaseraud_impl(w.data[0], freq_scaled, startphaseleft, fb, depth, stages, drywet); effect_phaseraud_impl(w.data[1], freq_scaled, startphaseright, fb, depth, stages, drywet); } return(w); }
public static WaveAudio Wahwah(WaveAudio wOriginal, double freq, double depth, double freqofs, double res) { double startphaseleft = 0; double startphaseright = startphaseleft + Math.PI; //note that left and right channels should start pi out of phase double freq_scaled = 2.0 * Math.PI * freq / (double)wOriginal.getSampleRate(); WaveAudio w = wOriginal.Clone(); if (w.getNumChannels() == 1) { effect_wahwahaud_impl(w.data[0], startphaseleft, freq_scaled, depth, freqofs, res); } else { effect_wahwahaud_impl(w.data[0], startphaseleft, freq_scaled, depth, freqofs, res); effect_wahwahaud_impl(w.data[1], startphaseright, freq_scaled, depth, freqofs, res); } return(w); }
public static WaveAudio Derivative(WaveAudio wOriginal) { WaveAudio w = wOriginal.Clone(); for (int ch = 0; ch < w.getNumChannels(); ch++) { for (int i = 0; i < w.data[ch].Length; i++) { if (i + 1 < w.data[ch].Length) { w.data[ch][i] = w.data[ch][i] - w.data[ch][i + 1]; } else { w.data[ch][i] = w.data[ch][i] - 0; } } } return(w); }
public static void propertytests() { // these aren't the best tests. WaveAudio w1 = new WaveAudio(44100, 2); asserteq(w1.data.Length, 2, "channels"); asserteq(w1.getNumChannels(), 2, "channels"); assert(w1.data[0] != null && w1.data[1] != null, "channels"); assert(w1.data[0].Length == 1 && w1.data[1].Length == 1, "004"); asserteq(w1.getSampleRate(), 44100, "005"); WaveAudio w1m = new WaveAudio(22050, 1); asserteq(w1m.data.Length, 1, "channels"); assert(w1m.data[0] != null, "channels"); asserteq(w1m.data[0].Length, 1, "004"); asserteq(w1m.getSampleRate(), 22050, "005"); // now set some properties w1m.LengthInSamples = 100; asserteq(w1m.data[0].Length, 100); asserteqf(w1m.LengthInSeconds, 100 / (double)w1m.getSampleRate(), 0.001); }
/// <summary> /// Find spectrum content of signal. Returns array, where each element represents energy at frequencies. /// For example, SpectrumContent(w, 8) returns 8 numbers. The first in the array is the amount of energy at low frequencies, and the last is the energy at highest frequencies. /// Note that FFT uses a power of 2 samples, and so all of the signal may not be used. /// </summary> /// <param name="w">Sound</param> /// <param name="nBins">Number of bins to return.</param> public static double[] SpectrumContent(WaveAudio w, int nBins) { if (w.getNumChannels() != 1) { throw new Exception("Only mono supported."); } double[] buffer; // FFT uses a power of 2 samples, so we might have to truncate. if (!isPowerOfTwo((uint)w.LengthInSamples)) { int nSize = (int)findNearestPowerOfTwo((uint)w.LengthInSamples); buffer = new double[nSize]; Array.Copy(w.data[0], buffer, nSize); } else { buffer = w.data[0]; } return(getSpectrumContent(buffer, nBins)); }
// These work by shifting the signal until it seems to correlate with itself. // In other words if the signal looks very similar to (signal shifted 200 samples) than the fundamental period is probably 200 samples // Note that the algorithm only works well when there's only one prominent fundamental. // This could be optimized by looking at the rate of change to determine a maximum without testing all periods. private static double[] detectPitchCalculation(WaveAudio w, double minHz, double maxHz, int nCandidates, int nResolution, PitchDetectAlgorithm algorithm) { // note that higher frequency means lower period int nLowPeriodInSamples = hzToPeriodInSamples(maxHz, w.getSampleRate()); int nHiPeriodInSamples = hzToPeriodInSamples(minHz, w.getSampleRate()); if (nHiPeriodInSamples <= nLowPeriodInSamples) { throw new Exception("Bad range for pitch detection."); } if (w.getNumChannels() != 1) { throw new Exception("Only mono supported."); } double[] samples = w.data[0]; if (samples.Length < nHiPeriodInSamples) { throw new Exception("Not enough samples."); } // both algorithms work in a similar way // they yield an array of data, and then we find the index at which the value is highest. double[] results = new double[nHiPeriodInSamples - nLowPeriodInSamples]; if (algorithm == PitchDetectAlgorithm.Amdf) { for (int period = nLowPeriodInSamples; period < nHiPeriodInSamples; period += nResolution) { double sum = 0; // for each sample, see how close it is to a sample n away. Then sum these. for (int i = 0; i < samples.Length - period; i++) { sum += Math.Abs(samples[i] - samples[i + period]); } double mean = sum / (double)samples.Length; mean *= -1; //somewhat of a hack. We are trying to find the minimum value, but our findBestCandidates finds the max. value. results[period - nLowPeriodInSamples] = mean; } } else if (algorithm == PitchDetectAlgorithm.Autocorrelation) { for (int period = nLowPeriodInSamples; period < nHiPeriodInSamples; period += nResolution) { double sum = 0; // for each sample, find correlation. (If they are far apart, small) for (int i = 0; i < samples.Length - period; i++) { sum += samples[i] * samples[i + period]; } double mean = sum / (double)samples.Length; results[period - nLowPeriodInSamples] = mean; } } // find the best indices int[] bestIndices = findBestCandidates(nCandidates, ref results); //note findBestCandidates modifies parameter // convert back to Hz double[] res = new double[nCandidates]; for (int i = 0; i < nCandidates; i++) { res[i] = periodInSamplesToHz(bestIndices[i] + nLowPeriodInSamples, w.getSampleRate()); } return(res); }
// Helper function. It's long and gross because either sound could be longer. // I could be more clever and use Math.Max / Min to have WaveAudio longer, WaveAudio shorter // , but at least now it is readable /// <summary> /// Element-wise combination of two audio clips. For example, adding, or modulation. /// </summary> internal static WaveAudio elementWiseCombination(WaveAudio w1, WaveAudio w2, ElementWiseCombinationFn fn) { if (w1.m_currentSampleRate != w2.m_currentSampleRate) { throw new Exception("Sample rates don't match"); } if (w1.getNumChannels() != w2.getNumChannels()) { throw new Exception("Number of channels don't match"); } WaveAudio newwave = new WaveAudio(w1.getSampleRate(), w1.getNumChannels()); newwave.LengthInSamples = Math.Max(w1.LengthInSamples, w2.LengthInSamples); double val; for (int ch = 0; ch < w1.getNumChannels(); ch++) { if (w1.LengthInSamples > w2.LengthInSamples) { for (int i = 0; i < w1.LengthInSamples; i++) { if (i >= w2.LengthInSamples) { val = fn(w1.data[ch][i], 0); } else { val = fn(w1.data[ch][i], w2.data[ch][i]); } if (val > 1.0) { val = 1.0; } else if (val < -1.0) { val = -1.0; } newwave.data[ch][i] = val; } } else { for (int i = 0; i < w2.LengthInSamples; i++) { if (i >= w1.LengthInSamples) { val = fn(0, w2.data[ch][i]); } else { val = fn(w1.data[ch][i], w2.data[ch][i]); } if (val > 1.0) { val = 1.0; } else if (val < -1.0) { val = -1.0; } newwave.data[ch][i] = val; } } } return(newwave); }
public static WaveAudio Vibrato(WaveAudio wave, double freq, double width) { if (width < 0) throw new ArgumentException("Factor must >= 0"); WaveAudio newwave = new WaveAudio(wave.getSampleRate(), wave.getNumChannels()); // do operation for all channels for (int i = 0; i < wave.getNumChannels(); i++) newwave.data[i] = vibratoChannel(wave.data[i], wave.getSampleRate(), width, freq); return newwave; }
static void iotests_perfile(string strFilename, int nBits, int nChannels, int nRate) { WaveAudio w01 = new WaveAudio(strFilename); asserteq(w01.getNumChannels(), nChannels); asserteq(w01.getSampleRate(), nRate, "011"); asserteqf(w01.LengthInSamples, 90725 * (nRate / 22050), 1.0, "012"); //note give 1.0 tolerance asserteqf(w01.LengthInSeconds, 4.1145124, "013"); asserteq(w01.data.Length, nChannels); asserteq(w01.data[0].Length, w01.LengthInSamples); for (int i = 0; i < nChannels; i++) asserteq(w01.data[i].Length, w01.LengthInSamples); // test converting to other rates / quality w01.SaveWaveFile("..\\..\\testout\\o_" + nRate + "_" + nBits + "_" + nRate + "_" + nChannels + ".wav", nBits); nBits = (nBits == 8) ? 16 : 8; w01.SaveWaveFile("..\\..\\testout\\ot_" + nRate + "_" + nBits + "_" + nRate + "_" + nChannels + ".wav", nBits); }
public static WaveAudio Tremolo(WaveAudio w, double tremfreq, double amp) { WaveAudio res = new WaveAudio(w.getSampleRate(), w.getNumChannels()); res.LengthInSamples = w.LengthInSamples; double tremeloFreqScale = 2.0 * Math.PI * tremfreq / (double)w.getSampleRate(); for (int ch=0;ch<w.data.Length; ch++) { for (int i = 0; i < w.data[ch].Length; i++) { double val = w.data[ch][i] * (1 + amp * Math.Sin(tremeloFreqScale * i)); if (val > 1.0) val = 1.0; else if (val < -1.0) val = -1.0; res.data[ch][i] = val; } } return res; }
// Test between 20Hz and 500Hz. This works very well, although it is hard to eliminate octave errors // This test will play original, then a sine at the frequency it detected. The two should line up. private static void pitchdetectwav(AudioPlayer pl, string strFilename, PitchDetection.PitchDetectAlgorithm algorithm) { string strInstdir = mediadirpitch; WaveAudio w = new WaveAudio(strInstdir + strFilename); if (w.getNumChannels() != 1) w.setNumChannels(1, true); double dfreq = PitchDetection.DetectPitch(w, 50,500,algorithm); WaveAudio testPitch = new Triangle(dfreq, 0.7).CreateWaveAudio(1.0); pl.Play(w); pl.Play(testPitch); }
// Helper function. It's long and gross because either sound could be longer. // I could be more clever and use Math.Max / Min to have WaveAudio longer, WaveAudio shorter // , but at least now it is readable /// <summary> /// Element-wise combination of two audio clips. For example, adding, or modulation. /// </summary> internal static WaveAudio elementWiseCombination(WaveAudio w1, WaveAudio w2, ElementWiseCombinationFn fn) { if (w1.m_currentSampleRate != w2.m_currentSampleRate) throw new Exception("Sample rates don't match"); if (w1.getNumChannels() != w2.getNumChannels()) throw new Exception("Number of channels don't match"); WaveAudio newwave = new WaveAudio(w1.getSampleRate(), w1.getNumChannels()); newwave.LengthInSamples = Math.Max(w1.LengthInSamples, w2.LengthInSamples); double val; for (int ch = 0; ch < w1.getNumChannels(); ch++) { if (w1.LengthInSamples > w2.LengthInSamples) { for (int i = 0; i < w1.LengthInSamples; i++) { if (i >= w2.LengthInSamples) val = fn(w1.data[ch][i], 0); else val = fn(w1.data[ch][i], w2.data[ch][i]); if (val > 1.0) val = 1.0; else if (val < -1.0) val = -1.0; newwave.data[ch][i] = val; } } else { for (int i = 0; i < w2.LengthInSamples; i++) { if (i >= w1.LengthInSamples) val = fn(0, w2.data[ch][i]); else val = fn(w1.data[ch][i], w2.data[ch][i]); if (val > 1.0) val = 1.0; else if (val < -1.0) val = -1.0; newwave.data[ch][i] = val; } } } return newwave; }