Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

[DiffSinger] Add more mel checks between vocoder and acoustic model #1053

Merged
merged 2 commits into from
Mar 28, 2024
Merged

[DiffSinger] Add more mel checks between vocoder and acoustic model #1053

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
b421d91
Add more mel checks between vocoder and acoustic model
yqzhishen Feb 26, 2024
93950e5
Add checks for `win_size` and `fft_size`
yqzhishen Feb 26, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
11 changes: 9 additions & 2 deletions OpenUtau.Core/DiffSinger/DiffSingerConfig.cs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -33,15 +33,22 @@ public class DsConfig {
public string linguistic;
public string pitch;
public string variance;
public int hop_size = 512;
public int sample_rate = 44100;
public bool predict_dur = true;
public bool predict_energy = true;
public bool predict_breathiness = true;
public bool predict_voicing = false;
public bool predict_tension = false;
public bool use_expr = false;
public bool use_note_rest = false;
public int sample_rate = 44100;
public int hop_size = 512;
public int win_size = 2048;
public int fft_size = 2048;
public int num_mel_bins = 128;
public double mel_fmin = 40;
public double mel_fmax = 16000;
public string mel_base = "10"; // or "e"
public string mel_scale = "slaney"; // or "htk"
public float frameMs(){
return 1000f * hop_size / sample_rate;
}
Expand Down
78 changes: 74 additions & 4 deletions OpenUtau.Core/DiffSinger/DiffSingerRenderer.cs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -134,12 +134,62 @@ float[] InvokeDiffsinger(RenderPhrase phrase, int depth, int speedup, Cancellati
}

var vocoder = singer.getVocoder();
//Vocoder and singer should have the same hop sizes and sample rates.
//mel specification validity checks
//mel base must be 10 or e
if (vocoder.mel_base != "10" && vocoder.mel_base != "e") {
throw new Exception(
$"Mel base must be \"10\" or \"e\", but got \"{vocoder.mel_base}\" from vocoder");
}
if (singer.dsConfig.mel_base != "10" && singer.dsConfig.mel_base != "e") {
throw new Exception(
$"Mel base must be \"10\" or \"e\", but got \"{singer.dsConfig.mel_base}\" from acoustic model");
}
//mel scale must be slaney or htk
if (vocoder.mel_scale != "slaney" && vocoder.mel_scale != "htk") {
throw new Exception(
$"Mel scale must be \"slaney\" or \"htk\", but got \"{vocoder.mel_scale}\" from vocoder");
}
if (singer.dsConfig.mel_scale != "slaney" && singer.dsConfig.mel_scale != "htk") {
throw new Exception(
$"Mel scale must be \"slaney\" or \"htk\", but got \"{vocoder.mel_scale}\" from acoustic model");
}
//mel specification matching checks
if(vocoder.sample_rate != singer.dsConfig.sample_rate) {
throw new Exception(
$"Vocoder and acoustic model has mismatching sample rate ({vocoder.sample_rate} != {singer.dsConfig.sample_rate})");
}
if(vocoder.hop_size != singer.dsConfig.hop_size){
throw new Exception($"Vocoder's hop size is {vocoder.hop_size}, but acoustic's hop size is {singer.dsConfig.hop_size}.");
throw new Exception(
$"Vocoder and acoustic model has mismatching hop size ({vocoder.hop_size} != {singer.dsConfig.hop_size})");
}
if(vocoder.win_size != singer.dsConfig.win_size){
throw new Exception(
$"Vocoder and acoustic model has mismatching win size ({vocoder.win_size} != {singer.dsConfig.win_size})");
}
if(vocoder.fft_size != singer.dsConfig.fft_size){
throw new Exception(
$"Vocoder and acoustic model has mismatching FFT size ({vocoder.fft_size} != {singer.dsConfig.fft_size})");
}
if (vocoder.num_mel_bins != singer.dsConfig.num_mel_bins) {
throw new Exception(
$"Vocoder and acoustic model has mismatching mel bins ({vocoder.num_mel_bins} != {singer.dsConfig.num_mel_bins})");
}
if (Math.Abs(vocoder.mel_fmin - singer.dsConfig.mel_fmin) > 1e-5) {
throw new Exception(
$"Vocoder and acoustic model has mismatching fmin ({vocoder.mel_fmin} != {singer.dsConfig.mel_fmin})");
}
if(vocoder.sample_rate != singer.dsConfig.sample_rate){
throw new Exception($"Vocoder's sample rate is {vocoder.sample_rate}, but acoustic's sample rate is {singer.dsConfig.sample_rate}.");
if (Math.Abs(vocoder.mel_fmax - singer.dsConfig.mel_fmax) > 1e-5) {
throw new Exception(
$"Vocoder and acoustic model has mismatching fmax ({vocoder.mel_fmax} != {singer.dsConfig.mel_fmax})");
}
// mismatching mel base can be transformed
// if (vocoder.mel_base != singer.dsConfig.mel_base) {
// throw new Exception(
// $"Vocoder and acoustic model has mismatching mel base ({vocoder.mel_base} != {singer.dsConfig.mel_base})");
// }
if (vocoder.mel_scale != singer.dsConfig.mel_scale) {
throw new Exception(
$"Vocoder and acoustic model has mismatching mel scale ({vocoder.mel_scale} != {singer.dsConfig.mel_scale})");
}

var acousticModel = singer.getAcousticSession();
Expand Down Expand Up @@ -301,6 +351,26 @@ float[] InvokeDiffsinger(RenderPhrase phrase, int depth, int speedup, Cancellati
var acousticOutputs = acousticModel.Run(acousticInputs);
mel = acousticOutputs.First().AsTensor<float>().Clone();
}
//mel transforms for different mel base
if (vocoder.mel_base != singer.dsConfig.mel_base) {
float k;
if (vocoder.mel_base == "e" && singer.dsConfig.mel_base == "10") {
k = 2.30259f;
}
else if (vocoder.mel_base == "10" && singer.dsConfig.mel_base == "e") {
k = 0.434294f;
} else {
// this should never happen
throw new Exception("This should never happen");
}
for (int b = 0; b < mel.Dimensions[0]; ++b) {
for (int t = 0; t < mel.Dimensions[1]; ++t) {
for (int c = 0; c < mel.Dimensions[2]; ++c) {
mel[b, t, c] *= k;
}
}
}
}
//vocoder
//waveform = session.run(['waveform'], {'mel': mel, 'f0': f0})[0]
var vocoderInputs = new List<NamedOnnxValue>();
Expand Down
16 changes: 14 additions & 2 deletions OpenUtau.Core/DiffSinger/DiffSingerVocoder.cs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -10,7 +10,13 @@ public class DsVocoder : IDisposable {

public int num_mel_bins => config.num_mel_bins;
public int hop_size => config.hop_size;
public int win_size => config.win_size;
public int fft_size => config.fft_size;
public int sample_rate => config.sample_rate;
public double mel_fmin => config.mel_fmin;
public double mel_fmax => config.mel_fmax;
public string mel_base => config.mel_base;
public string mel_scale => config.mel_scale;

//Get vocoder by package name
public DsVocoder(string name) {
Expand Down Expand Up @@ -54,8 +60,14 @@ public void Dispose() {
public class DsVocoderConfig {
public string name = "vocoder";
public string model = "model.onnx";
public int num_mel_bins = 128;
public int hop_size = 512;
public int sample_rate = 44100;
public int hop_size = 512;
public int win_size = 2048;
public int fft_size = 2048;
public int num_mel_bins = 128;
public double mel_fmin = 40;
public double mel_fmax = 16000;
public string mel_base = "10"; // or "e"
public string mel_scale = "slaney"; // or "htk"
}
}