Spaces:
Running
Running
Update app.py
Browse files
app.py
CHANGED
@@ -5,9 +5,9 @@ import librosa
|
|
5 |
from modules.commons import build_model, load_checkpoint, recursive_munch
|
6 |
import yaml
|
7 |
from hf_utils import load_custom_model_from_hf
|
8 |
-
import spaces
|
9 |
import numpy as np
|
10 |
from pydub import AudioSegment
|
|
|
11 |
|
12 |
# Load model and configuration
|
13 |
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
|
@@ -33,8 +33,9 @@ model.cfm.estimator.setup_caches(max_batch_size=1, max_seq_length=8192)
|
|
33 |
# Load additional modules
|
34 |
from modules.campplus.DTDNN import CAMPPlus
|
35 |
|
|
|
36 |
campplus_model = CAMPPlus(feat_dim=80, embedding_size=192)
|
37 |
-
campplus_model.load_state_dict(torch.load(
|
38 |
campplus_model.eval()
|
39 |
campplus_model.to(device)
|
40 |
|
@@ -50,6 +51,14 @@ hift_gen.load_state_dict(torch.load(hift_checkpoint_path, map_location='cpu'))
|
|
50 |
hift_gen.eval()
|
51 |
hift_gen.to(device)
|
52 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
53 |
speech_tokenizer_type = config['model_params']['speech_tokenizer'].get('type', 'cosyvoice')
|
54 |
if speech_tokenizer_type == 'cosyvoice':
|
55 |
from modules.cosyvoice_tokenizer.frontend import CosyVoiceFrontEnd
|
@@ -69,6 +78,7 @@ elif speech_tokenizer_type == 'facodec':
|
|
69 |
codec_encoder[key].load_state_dict(ckpt_params[key], strict=False)
|
70 |
_ = [codec_encoder[key].eval() for key in codec_encoder]
|
71 |
_ = [codec_encoder[key].to(device) for key in codec_encoder]
|
|
|
72 |
# Generate mel spectrograms
|
73 |
mel_fn_args = {
|
74 |
"n_fft": config['preprocess_params']['spect_params']['n_fft'],
|
@@ -80,13 +90,24 @@ mel_fn_args = {
|
|
80 |
"fmax": 8000,
|
81 |
"center": False
|
82 |
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
83 |
from modules.audio import mel_spectrogram
|
84 |
|
85 |
to_mel = lambda x: mel_spectrogram(x, **mel_fn_args)
|
|
|
86 |
|
87 |
# f0 conditioned model
|
88 |
dit_checkpoint_path, dit_config_path = load_custom_model_from_hf("Plachta/Seed-VC",
|
89 |
-
"
|
90 |
"config_dit_mel_seed_facodec_small_wavenet_f0.yml")
|
91 |
|
92 |
config = yaml.safe_load(open(dit_config_path, 'r'))
|
@@ -114,8 +135,8 @@ def adjust_f0_semitones(f0_sequence, n_semitones):
|
|
114 |
return f0_sequence * factor
|
115 |
|
116 |
def crossfade(chunk1, chunk2, overlap):
|
117 |
-
fade_out = np.
|
118 |
-
fade_in = np.
|
119 |
chunk2[:overlap] = chunk2[:overlap] * fade_in + chunk1[-overlap:] * fade_out
|
120 |
return chunk2
|
121 |
|
@@ -123,14 +144,14 @@ def crossfade(chunk1, chunk2, overlap):
|
|
123 |
max_context_window = sr // hop_length * 30
|
124 |
overlap_frame_len = 64
|
125 |
overlap_wave_len = overlap_frame_len * hop_length
|
126 |
-
max_wave_len_per_chunk = 24000 * 20
|
127 |
bitrate = "320k"
|
128 |
|
129 |
-
@spaces.GPU
|
130 |
@torch.no_grad()
|
131 |
@torch.inference_mode()
|
|
|
132 |
def voice_conversion(source, target, diffusion_steps, length_adjust, inference_cfg_rate, n_quantizers, f0_condition, auto_f0_adjust, pitch_shift):
|
133 |
inference_module = model if not f0_condition else model_f0
|
|
|
134 |
# Load audio
|
135 |
source_audio = librosa.load(source, sr=sr)[0]
|
136 |
ref_audio = librosa.load(target, sr=sr)[0]
|
@@ -150,6 +171,7 @@ def voice_conversion(source, target, diffusion_steps, length_adjust, inference_c
|
|
150 |
elif speech_tokenizer_type == 'facodec':
|
151 |
converted_waves_24k = torchaudio.functional.resample(source_audio, sr, 24000)
|
152 |
waves_input = converted_waves_24k.unsqueeze(1)
|
|
|
153 |
wave_input_chunks = [
|
154 |
waves_input[..., i:i + max_wave_len_per_chunk] for i in range(0, waves_input.size(-1), max_wave_len_per_chunk)
|
155 |
]
|
@@ -180,8 +202,8 @@ def voice_conversion(source, target, diffusion_steps, length_adjust, inference_c
|
|
180 |
)
|
181 |
S_ori = torch.cat([codes[1], codes[0]], dim=1)
|
182 |
|
183 |
-
mel =
|
184 |
-
mel2 =
|
185 |
|
186 |
target_lengths = torch.LongTensor([int(mel.size(2) * length_adjust)]).to(mel.device)
|
187 |
target2_lengths = torch.LongTensor([mel2.size(2)]).to(mel2.device)
|
@@ -194,8 +216,8 @@ def voice_conversion(source, target, diffusion_steps, length_adjust, inference_c
|
|
194 |
style2 = campplus_model(feat2.unsqueeze(0))
|
195 |
|
196 |
if f0_condition:
|
197 |
-
waves_16k = torchaudio.functional.resample(waves_24k,
|
198 |
-
converted_waves_16k = torchaudio.functional.resample(converted_waves_24k,
|
199 |
F0_ori = rmvpe.infer_from_audio(waves_16k[0], thred=0.03)
|
200 |
F0_alt = rmvpe.infer_from_audio(converted_waves_16k[0], thred=0.03)
|
201 |
|
@@ -244,7 +266,10 @@ def voice_conversion(source, target, diffusion_steps, length_adjust, inference_c
|
|
244 |
mel2, style2, None, diffusion_steps,
|
245 |
inference_cfg_rate=inference_cfg_rate)
|
246 |
vc_target = vc_target[:, :, mel2.size(-1):]
|
247 |
-
|
|
|
|
|
|
|
248 |
if processed_frames == 0:
|
249 |
if is_last_chunk:
|
250 |
output_wave = vc_wave[0].cpu().numpy()
|
@@ -254,7 +279,7 @@ def voice_conversion(source, target, diffusion_steps, length_adjust, inference_c
|
|
254 |
output_wave.tobytes(), frame_rate=sr,
|
255 |
sample_width=output_wave.dtype.itemsize, channels=1
|
256 |
).export(format="mp3", bitrate=bitrate).read()
|
257 |
-
yield mp3_bytes
|
258 |
break
|
259 |
output_wave = vc_wave[0, :-overlap_wave_len].cpu().numpy()
|
260 |
generated_wave_chunks.append(output_wave)
|
@@ -265,7 +290,7 @@ def voice_conversion(source, target, diffusion_steps, length_adjust, inference_c
|
|
265 |
output_wave.tobytes(), frame_rate=sr,
|
266 |
sample_width=output_wave.dtype.itemsize, channels=1
|
267 |
).export(format="mp3", bitrate=bitrate).read()
|
268 |
-
yield mp3_bytes
|
269 |
elif is_last_chunk:
|
270 |
output_wave = crossfade(previous_chunk.cpu().numpy(), vc_wave[0].cpu().numpy(), overlap_wave_len)
|
271 |
generated_wave_chunks.append(output_wave)
|
@@ -275,7 +300,7 @@ def voice_conversion(source, target, diffusion_steps, length_adjust, inference_c
|
|
275 |
output_wave.tobytes(), frame_rate=sr,
|
276 |
sample_width=output_wave.dtype.itemsize, channels=1
|
277 |
).export(format="mp3", bitrate=bitrate).read()
|
278 |
-
yield mp3_bytes
|
279 |
break
|
280 |
else:
|
281 |
output_wave = crossfade(previous_chunk.cpu().numpy(), vc_wave[0, :-overlap_wave_len].cpu().numpy(), overlap_wave_len)
|
@@ -287,7 +312,7 @@ def voice_conversion(source, target, diffusion_steps, length_adjust, inference_c
|
|
287 |
output_wave.tobytes(), frame_rate=sr,
|
288 |
sample_width=output_wave.dtype.itemsize, channels=1
|
289 |
).export(format="mp3", bitrate=bitrate).read()
|
290 |
-
yield mp3_bytes
|
291 |
|
292 |
|
293 |
if __name__ == "__main__":
|
@@ -308,10 +333,15 @@ if __name__ == "__main__":
|
|
308 |
]
|
309 |
|
310 |
examples = [["examples/source/yae_0.wav", "examples/reference/dingzhen_0.wav", 25, 1.0, 0.7, 1, False, True, 0],
|
311 |
-
["examples/source/
|
312 |
-
|
|
|
|
|
|
|
|
|
313 |
|
314 |
-
outputs = gr.Audio(label="Output Audio", streaming=True, format='mp3')
|
|
|
315 |
|
316 |
gr.Interface(fn=voice_conversion,
|
317 |
description=description,
|
@@ -320,4 +350,4 @@ if __name__ == "__main__":
|
|
320 |
title="Seed Voice Conversion",
|
321 |
examples=examples,
|
322 |
cache_examples=False,
|
323 |
-
).launch()
|
|
|
5 |
from modules.commons import build_model, load_checkpoint, recursive_munch
|
6 |
import yaml
|
7 |
from hf_utils import load_custom_model_from_hf
|
|
|
8 |
import numpy as np
|
9 |
from pydub import AudioSegment
|
10 |
+
import spaces
|
11 |
|
12 |
# Load model and configuration
|
13 |
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
|
|
|
33 |
# Load additional modules
|
34 |
from modules.campplus.DTDNN import CAMPPlus
|
35 |
|
36 |
+
campplus_ckpt_path = load_custom_model_from_hf("funasr/campplus", "campplus_cn_common.bin", config_filename=None)
|
37 |
campplus_model = CAMPPlus(feat_dim=80, embedding_size=192)
|
38 |
+
campplus_model.load_state_dict(torch.load(campplus_ckpt_path, map_location="cpu"))
|
39 |
campplus_model.eval()
|
40 |
campplus_model.to(device)
|
41 |
|
|
|
51 |
hift_gen.eval()
|
52 |
hift_gen.to(device)
|
53 |
|
54 |
+
from modules.bigvgan import bigvgan
|
55 |
+
|
56 |
+
bigvgan_model = bigvgan.BigVGAN.from_pretrained('nvidia/bigvgan_v2_22khz_80band_256x', use_cuda_kernel=False)
|
57 |
+
|
58 |
+
# remove weight norm in the model and set to eval mode
|
59 |
+
bigvgan_model.remove_weight_norm()
|
60 |
+
bigvgan_model = bigvgan_model.eval().to(device)
|
61 |
+
|
62 |
speech_tokenizer_type = config['model_params']['speech_tokenizer'].get('type', 'cosyvoice')
|
63 |
if speech_tokenizer_type == 'cosyvoice':
|
64 |
from modules.cosyvoice_tokenizer.frontend import CosyVoiceFrontEnd
|
|
|
78 |
codec_encoder[key].load_state_dict(ckpt_params[key], strict=False)
|
79 |
_ = [codec_encoder[key].eval() for key in codec_encoder]
|
80 |
_ = [codec_encoder[key].to(device) for key in codec_encoder]
|
81 |
+
|
82 |
# Generate mel spectrograms
|
83 |
mel_fn_args = {
|
84 |
"n_fft": config['preprocess_params']['spect_params']['n_fft'],
|
|
|
90 |
"fmax": 8000,
|
91 |
"center": False
|
92 |
}
|
93 |
+
mel_fn_args_f0 = {
|
94 |
+
"n_fft": config['preprocess_params']['spect_params']['n_fft'],
|
95 |
+
"win_size": config['preprocess_params']['spect_params']['win_length'],
|
96 |
+
"hop_size": config['preprocess_params']['spect_params']['hop_length'],
|
97 |
+
"num_mels": config['preprocess_params']['spect_params']['n_mels'],
|
98 |
+
"sampling_rate": sr,
|
99 |
+
"fmin": 0,
|
100 |
+
"fmax": None,
|
101 |
+
"center": False
|
102 |
+
}
|
103 |
from modules.audio import mel_spectrogram
|
104 |
|
105 |
to_mel = lambda x: mel_spectrogram(x, **mel_fn_args)
|
106 |
+
to_mel_f0 = lambda x: mel_spectrogram(x, **mel_fn_args_f0)
|
107 |
|
108 |
# f0 conditioned model
|
109 |
dit_checkpoint_path, dit_config_path = load_custom_model_from_hf("Plachta/Seed-VC",
|
110 |
+
"DiT_seed_v2_uvit_facodec_small_wavenet_f0_bigvgan_pruned.pth",
|
111 |
"config_dit_mel_seed_facodec_small_wavenet_f0.yml")
|
112 |
|
113 |
config = yaml.safe_load(open(dit_config_path, 'r'))
|
|
|
135 |
return f0_sequence * factor
|
136 |
|
137 |
def crossfade(chunk1, chunk2, overlap):
|
138 |
+
fade_out = np.linspace(1, 0, overlap)
|
139 |
+
fade_in = np.linspace(0, 1, overlap)
|
140 |
chunk2[:overlap] = chunk2[:overlap] * fade_in + chunk1[-overlap:] * fade_out
|
141 |
return chunk2
|
142 |
|
|
|
144 |
max_context_window = sr // hop_length * 30
|
145 |
overlap_frame_len = 64
|
146 |
overlap_wave_len = overlap_frame_len * hop_length
|
|
|
147 |
bitrate = "320k"
|
148 |
|
|
|
149 |
@torch.no_grad()
|
150 |
@torch.inference_mode()
|
151 |
+
@spaces.GPU
|
152 |
def voice_conversion(source, target, diffusion_steps, length_adjust, inference_cfg_rate, n_quantizers, f0_condition, auto_f0_adjust, pitch_shift):
|
153 |
inference_module = model if not f0_condition else model_f0
|
154 |
+
mel_fn = to_mel if not f0_condition else to_mel_f0
|
155 |
# Load audio
|
156 |
source_audio = librosa.load(source, sr=sr)[0]
|
157 |
ref_audio = librosa.load(target, sr=sr)[0]
|
|
|
171 |
elif speech_tokenizer_type == 'facodec':
|
172 |
converted_waves_24k = torchaudio.functional.resample(source_audio, sr, 24000)
|
173 |
waves_input = converted_waves_24k.unsqueeze(1)
|
174 |
+
max_wave_len_per_chunk = 24000 * 20
|
175 |
wave_input_chunks = [
|
176 |
waves_input[..., i:i + max_wave_len_per_chunk] for i in range(0, waves_input.size(-1), max_wave_len_per_chunk)
|
177 |
]
|
|
|
202 |
)
|
203 |
S_ori = torch.cat([codes[1], codes[0]], dim=1)
|
204 |
|
205 |
+
mel = mel_fn(source_audio.to(device).float())
|
206 |
+
mel2 = mel_fn(ref_audio.to(device).float())
|
207 |
|
208 |
target_lengths = torch.LongTensor([int(mel.size(2) * length_adjust)]).to(mel.device)
|
209 |
target2_lengths = torch.LongTensor([mel2.size(2)]).to(mel2.device)
|
|
|
216 |
style2 = campplus_model(feat2.unsqueeze(0))
|
217 |
|
218 |
if f0_condition:
|
219 |
+
waves_16k = torchaudio.functional.resample(waves_24k, 24000, 16000)
|
220 |
+
converted_waves_16k = torchaudio.functional.resample(converted_waves_24k, 24000, 16000)
|
221 |
F0_ori = rmvpe.infer_from_audio(waves_16k[0], thred=0.03)
|
222 |
F0_alt = rmvpe.infer_from_audio(converted_waves_16k[0], thred=0.03)
|
223 |
|
|
|
266 |
mel2, style2, None, diffusion_steps,
|
267 |
inference_cfg_rate=inference_cfg_rate)
|
268 |
vc_target = vc_target[:, :, mel2.size(-1):]
|
269 |
+
if not f0_condition:
|
270 |
+
vc_wave = hift_gen.inference(vc_target, f0=None)
|
271 |
+
else:
|
272 |
+
vc_wave = bigvgan_model(vc_target)[0]
|
273 |
if processed_frames == 0:
|
274 |
if is_last_chunk:
|
275 |
output_wave = vc_wave[0].cpu().numpy()
|
|
|
279 |
output_wave.tobytes(), frame_rate=sr,
|
280 |
sample_width=output_wave.dtype.itemsize, channels=1
|
281 |
).export(format="mp3", bitrate=bitrate).read()
|
282 |
+
yield mp3_bytes, (sr, np.concatenate(generated_wave_chunks))
|
283 |
break
|
284 |
output_wave = vc_wave[0, :-overlap_wave_len].cpu().numpy()
|
285 |
generated_wave_chunks.append(output_wave)
|
|
|
290 |
output_wave.tobytes(), frame_rate=sr,
|
291 |
sample_width=output_wave.dtype.itemsize, channels=1
|
292 |
).export(format="mp3", bitrate=bitrate).read()
|
293 |
+
yield mp3_bytes, None
|
294 |
elif is_last_chunk:
|
295 |
output_wave = crossfade(previous_chunk.cpu().numpy(), vc_wave[0].cpu().numpy(), overlap_wave_len)
|
296 |
generated_wave_chunks.append(output_wave)
|
|
|
300 |
output_wave.tobytes(), frame_rate=sr,
|
301 |
sample_width=output_wave.dtype.itemsize, channels=1
|
302 |
).export(format="mp3", bitrate=bitrate).read()
|
303 |
+
yield mp3_bytes, (sr, np.concatenate(generated_wave_chunks))
|
304 |
break
|
305 |
else:
|
306 |
output_wave = crossfade(previous_chunk.cpu().numpy(), vc_wave[0, :-overlap_wave_len].cpu().numpy(), overlap_wave_len)
|
|
|
312 |
output_wave.tobytes(), frame_rate=sr,
|
313 |
sample_width=output_wave.dtype.itemsize, channels=1
|
314 |
).export(format="mp3", bitrate=bitrate).read()
|
315 |
+
yield mp3_bytes, None
|
316 |
|
317 |
|
318 |
if __name__ == "__main__":
|
|
|
333 |
]
|
334 |
|
335 |
examples = [["examples/source/yae_0.wav", "examples/reference/dingzhen_0.wav", 25, 1.0, 0.7, 1, False, True, 0],
|
336 |
+
["examples/source/jay_0.wav", "examples/reference/azuma_0.wav", 25, 1.0, 0.7, 1, True, True, 0],
|
337 |
+
["examples/source/Wiz Khalifa,Charlie Puth - See You Again [vocals]_[cut_28sec].wav",
|
338 |
+
"examples/reference/teio_0.wav", 100, 1.0, 0.7, 3, True, False, 0],
|
339 |
+
["examples/source/TECHNOPOLIS - 2085 [vocals]_[cut_14sec].wav",
|
340 |
+
"examples/reference/trump_0.wav", 50, 1.0, 0.7, 3, True, False, -12],
|
341 |
+
]
|
342 |
|
343 |
+
outputs = [gr.Audio(label="Stream Output Audio", streaming=True, format='mp3'),
|
344 |
+
gr.Audio(label="Full Output Audio", streaming=False, format='wav')]
|
345 |
|
346 |
gr.Interface(fn=voice_conversion,
|
347 |
description=description,
|
|
|
350 |
title="Seed Voice Conversion",
|
351 |
examples=examples,
|
352 |
cache_examples=False,
|
353 |
+
).launch()
|