PublicMirrors/voice-changer
1
0
Fork 0
mirror of https://github.com/w-okada/voice-changer.git synced 2025-01-23 21:45:00 +03:00
Code Issues Actions Packages Projects Releases Wiki Activity

WIP:crepe

Browse Source
This commit is contained in:
w-okada 2023-07-15 05:57:20 +09:00
parent 8339e72ef5
commit 485a747d55
3 changed files with 26 additions and 27 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

45
server/voice_changer/DiffusionSVC/pitchExtractor/CrepePitchExtractor.py
View File

@ -11,30 +11,26 @@ class CrepePitchExtractor(PitchExtractor):
def __init__(self): def __init__(self):
super().__init__() super().__init__()
self.pitchExtractorType: PitchExtractorType = "crepe" self.pitchExtractorType: PitchExtractorType = "crepe"
self.f0_min = 50
self.f0_max = 1100
self.sapmle_rate = 16000
self.uv_interp = True
if torch.cuda.is_available(): if torch.cuda.is_available():
self.device = torch.device("cuda:" + str(torch.cuda.current_device())) self.device = torch.device("cuda:" + str(torch.cuda.current_device()))
else: else:
self.device = torch.device("cpu") self.device = torch.device("cpu")
def extract(self, audio, pitchf, f0_up_key, sr, window, silence_front=0): def extract(self, audio: torch.Tensor, pitch, f0_up_key, window, silence_front=0):
n_frames = int(len(audio) // window) + 1 start_frame = int(silence_front * self.sapmle_rate / window)
start_frame = int(silence_front * sr / window) real_silence_front = start_frame * window / self.sapmle_rate
real_silence_front = start_frame * window / sr audio = audio[int(np.round(real_silence_front * self.sapmle_rate)):]
silence_front_offset = int(np.round(real_silence_front * sr))
audio = audio[silence_front_offset:]
f0_min = 50
f0_max = 1100
f0_mel_min = 1127 * np.log(1 + f0_min / 700)
f0_mel_max = 1127 * np.log(1 + f0_max / 700)
f0, pd = torchcrepe.predict( f0, pd = torchcrepe.predict(
audio.unsqueeze(0), audio.unsqueeze(0),
sr, self.sapmle_rate,
hop_length=window, hop_length=window,
fmin=f0_min, fmin=self.f0_min,
fmax=f0_max, fmax=self.f0_max,
# model="tiny", # model="tiny",
model="full", model="full",
batch_size=256, batch_size=256,
@ -46,14 +42,15 @@ class CrepePitchExtractor(PitchExtractor):
pd = torchcrepe.filter.median(pd, 3) pd = torchcrepe.filter.median(pd, 3)
f0[pd < 0.1] = 0 f0[pd < 0.1] = 0
f0 = f0.squeeze() f0 = f0.squeeze()
pitch[-f0.shape[0]:] = f0.cpu()[:pitch.shape[0]]
f0 = pitch
f0 *= pow(2, f0_up_key / 12) if self.uv_interp:
pitchf[-f0.shape[0]:] = f0.detach().cpu().numpy()[:pitchf.shape[0]] uv = f0 == 0
f0bak = pitchf.copy() if len(f0[~uv]) > 0:
f0_mel = 1127.0 * np.log(1.0 + f0bak / 700.0) f0[uv] = np.interp(np.where(uv)[0], np.where(~uv)[0], f0[~uv])
f0_mel = np.clip( f0[f0 < self.f0_min] = self.f0_min
(f0_mel - f0_mel_min) * 254.0 / (f0_mel_max - f0_mel_min) + 1.0, 1.0, 255.0
)
pitch_coarse = f0_mel.astype(int)
return pitch_coarse, pitchf f0 = f0 * 2 ** (float(f0_up_key) / 12)
return f0

5
server/voice_changer/DiffusionSVC/pitchExtractor/PitchExtractorManager.py
View File

@ -1,5 +1,6 @@
from typing import Protocol from typing import Protocol
from const import PitchExtractorType from const import PitchExtractorType
from voice_changer.DiffusionSVC.pitchExtractor.CrepePitchExtractor import CrepePitchExtractor
from voice_changer.DiffusionSVC.pitchExtractor.DioPitchExtractor import DioPitchExtractor from voice_changer.DiffusionSVC.pitchExtractor.DioPitchExtractor import DioPitchExtractor
from voice_changer.DiffusionSVC.pitchExtractor.HarvestPitchExtractor import HarvestPitchExtractor from voice_changer.DiffusionSVC.pitchExtractor.HarvestPitchExtractor import HarvestPitchExtractor
from voice_changer.DiffusionSVC.pitchExtractor.PitchExtractor import PitchExtractor from voice_changer.DiffusionSVC.pitchExtractor.PitchExtractor import PitchExtractor
@ -29,8 +30,8 @@ class PitchExtractorManager(Protocol):
return HarvestPitchExtractor() return HarvestPitchExtractor()
elif pitchExtractorType == "dio": elif pitchExtractorType == "dio":
return DioPitchExtractor() return DioPitchExtractor()
# elif pitchExtractorType == "crepe": elif pitchExtractorType == "crepe":
# return CrepePitchExtractor() return CrepePitchExtractor()
# elif pitchExtractorType == "crepe_tiny": # elif pitchExtractorType == "crepe_tiny":
# return CrepeOnnxPitchExtractor(pitchExtractorType, cls.params.crepe_onnx_tiny, gpu) # return CrepeOnnxPitchExtractor(pitchExtractorType, cls.params.crepe_onnx_tiny, gpu)
# elif pitchExtractorType == "crepe_full": # elif pitchExtractorType == "crepe_full":

1
server/voice_changer/common/VolumeExtractor.py
View File

@ -2,6 +2,7 @@ import numpy as np
import torch import torch
import torch.nn as nn import torch.nn as nn
class VolumeExtractor: class VolumeExtractor:
def __init__(self, hop_size: float): def __init__(self, hop_size: float):