public ConvertingSampleProvider(IWaveProvider waveProvider, WaveFormat targetFormat)
{
if (waveProvider.WaveFormat.SampleRate != targetFormat.SampleRate)
{
if (waveProvider.WaveFormat.Encoding == WaveFormatEncoding.IeeeFloat)
{
provider = ((resampler = new MediaFoundationResampler(waveProvider, targetFormat)) as MediaFoundationResampler).ToSampleProvider();
}
else
{
provider = ((resampler = new MediaFoundationResampler(new Wave16ToFloatProvider(waveProvider), targetFormat)) as MediaFoundationResampler).ToSampleProvider();
}
}
else
{
provider = waveProvider.ToSampleProvider();
}
if (provider.WaveFormat.Channels != targetFormat.Channels)
{
if (targetFormat.Channels == 1 & provider.WaveFormat.Channels > 1)
{
provider = provider.ToMono();
}
else if (targetFormat.Channels == 2)
{
provider = provider.ToStereo();
}
else
{
throw new InvalidWaveFormatException($"Couldn´t find a suitable conversion from {provider.WaveFormat.Channels} to {targetFormat.Channels} Channels");
}
}
}
public ISampleProvider SupplySample(bool stereo, bool loop, ISampleProvider input)
{
if (stereo)
{
input.ToStereo();
}
else
{
input.ToMono();
}
return(input);
}
public SampleProvider(string folderPath, ISampleProvider input, SourceLocation location)
{
this.hrtf = new HRTF(folderPath);
this.hrtf.Init();
this.binSyn = new BinauralSynthesis(currInput, dataOutput);
this.location = location;
// convert stereo input to mono
if (input.WaveFormat.Channels != 1)
{
this.input = input.ToMono();
}
else
{
this.input = input;
}
WaveFormat = input.ToStereo().WaveFormat;
}
public async ValueTask <VoiceAnalysisResult> Analyze(VowelClassifierType classifierType, ISampleProvider provider)
{
// TODO: あまりにも無駄な再計算が多い & 並列化しろ
const int vowelWindowSize = 2048;
const int pitchWindowSize = 1024;
provider = provider.ToMono();
var sampleRate = provider.WaveFormat.SampleRate;
var blocks = new List <NoteBlockModel>();
var classifier = await this._vowelClassifierTasks[classifierType].ConfigureAwait(false);
var samples = new float[Logics.AnalysisUnit];
for (var unitCount = 0; ; unitCount++)
{
// 4096 サンプルを読み込み
for (var readSamples = 0; readSamples < samples.Length;)
{
var count = provider.Read(samples, readSamples, samples.Length - readSamples);
if (count == 0)
{
return(new VoiceAnalysisResult(blocks.ToImmutableArray(), unitCount));
}
readSamples += count;
}
var maxPower = 0f;
foreach (var x in samples)
{
if (x > maxPower)
{
maxPower = x;
}
}
// 音量小さすぎ
if (maxPower < 0.15)
{
continue;
}
// 512 ずつずらしながら母音認識
var vowelCandidates = new int[(int)VowelType.Other + 1];
for (var offset = 0; offset <= Logics.AnalysisUnit - vowelWindowSize; offset += 512)
{
vowelCandidates[(int)classifier.Decide(new ReadOnlySpan <float>(samples, offset, vowelWindowSize), sampleRate)]++;
}
var vowelCandidate = default(VowelType?);
var maxNumOfVotes = 0;
for (var j = 0; j < vowelCandidates.Length; j++)
{
if (vowelCandidates[j] > maxNumOfVotes)
{
maxNumOfVotes = vowelCandidates[j];
vowelCandidate = (VowelType)j;
}
else if (vowelCandidates[j] == maxNumOfVotes)
{
vowelCandidate = null;
}
}
// 母音が定まらなかったので、終了
if (!vowelCandidate.HasValue || vowelCandidate.Value == VowelType.Other)
{
continue;
}
// 512 ずつずらしながらピッチ検出
const int pitchOffsetDelta = 512;
var basicFreqs = new List <double>((Logics.AnalysisUnit - pitchOffsetDelta) / pitchOffsetDelta);
for (var offset = 0; offset <= Logics.AnalysisUnit - pitchWindowSize; offset += pitchOffsetDelta)
{
var f = McLeodPitchMethod.EstimateFundamentalFrequency(
sampleRate,
new ReadOnlySpan <float>(samples, offset, pitchWindowSize)
);
if (f.HasValue)
{
basicFreqs.Add(f.Value);
}
}
// ピッチ検出に失敗したので終了
if (basicFreqs.Count == 0)
{
continue;
}
basicFreqs.Sort();
var basicFreq = basicFreqs[basicFreqs.Count / 2]; // 中央値
var noteNum = CommonUtils.HzToMidiNote(basicFreq);
var block = new NoteBlockModel(unitCount, noteNum, vowelCandidate.Value);
if (blocks.Count == 0 || !blocks[blocks.Count - 1].MergeIfPossible(block))
{
blocks.Add(block);
}
}
}