Spaces:
Running
Running
zhzluke96
commited on
Commit
·
9d9fe0d
1
Parent(s):
ba0472f
update
Browse files- .env.webui +2 -1
- modules/ChatTTS/ChatTTS/core.py +14 -8
- modules/Enhancer/ResembleEnhance.py +4 -5
- modules/api/api_setup.py +12 -1
- modules/webui/css/style.css +2 -2
- modules/webui/webui_utils.py +8 -4
.env.webui
CHANGED
|
@@ -9,8 +9,9 @@ AUTH=
|
|
| 9 |
|
| 10 |
# Model inference options
|
| 11 |
HALF=True
|
| 12 |
-
OFF_TQDM=
|
| 13 |
DEBUG_GENERATE=True
|
|
|
|
| 14 |
|
| 15 |
# Text-to-Speech (TTS) configuration
|
| 16 |
TTS_MAX_LEN=1000
|
|
|
|
| 9 |
|
| 10 |
# Model inference options
|
| 11 |
HALF=True
|
| 12 |
+
OFF_TQDM=
|
| 13 |
DEBUG_GENERATE=True
|
| 14 |
+
PRELOAD_MODELS=True
|
| 15 |
|
| 16 |
# Text-to-Speech (TTS) configuration
|
| 17 |
TTS_MAX_LEN=1000
|
modules/ChatTTS/ChatTTS/core.py
CHANGED
|
@@ -112,6 +112,8 @@ class Chat:
|
|
| 112 |
dtype_gpt = dtype_gpt or dtype
|
| 113 |
dtype_decoder = dtype_decoder or dtype
|
| 114 |
|
|
|
|
|
|
|
| 115 |
if vocos_config_path:
|
| 116 |
vocos = (
|
| 117 |
Vocos.from_hparams(vocos_config_path)
|
|
@@ -119,7 +121,9 @@ class Chat:
|
|
| 119 |
.eval()
|
| 120 |
)
|
| 121 |
assert vocos_ckpt_path, "vocos_ckpt_path should not be None"
|
| 122 |
-
vocos.load_state_dict(
|
|
|
|
|
|
|
| 123 |
self.pretrain_models["vocos"] = vocos
|
| 124 |
self.logger.log(logging.INFO, "vocos loaded.")
|
| 125 |
|
|
@@ -127,7 +131,7 @@ class Chat:
|
|
| 127 |
cfg = OmegaConf.load(dvae_config_path)
|
| 128 |
dvae = DVAE(**cfg).to(device=device, dtype=dtype_dvae).eval()
|
| 129 |
assert dvae_ckpt_path, "dvae_ckpt_path should not be None"
|
| 130 |
-
dvae.load_state_dict(torch.load(dvae_ckpt_path, map_location=
|
| 131 |
self.pretrain_models["dvae"] = dvae
|
| 132 |
self.logger.log(logging.INFO, "dvae loaded.")
|
| 133 |
|
|
@@ -135,7 +139,7 @@ class Chat:
|
|
| 135 |
cfg = OmegaConf.load(gpt_config_path)
|
| 136 |
gpt = GPT_warpper(**cfg).to(device=device, dtype=dtype_gpt).eval()
|
| 137 |
assert gpt_ckpt_path, "gpt_ckpt_path should not be None"
|
| 138 |
-
gpt.load_state_dict(torch.load(gpt_ckpt_path, map_location=
|
| 139 |
if compile and "cuda" in str(device):
|
| 140 |
self.logger.info("compile gpt model")
|
| 141 |
gpt.gpt.forward = torch.compile(
|
|
@@ -146,21 +150,23 @@ class Chat:
|
|
| 146 |
assert os.path.exists(
|
| 147 |
spk_stat_path
|
| 148 |
), f"Missing spk_stat.pt: {spk_stat_path}"
|
| 149 |
-
self.pretrain_models["spk_stat"] = torch.load(
|
| 150 |
-
|
| 151 |
-
)
|
| 152 |
self.logger.log(logging.INFO, "gpt loaded.")
|
| 153 |
|
| 154 |
if decoder_config_path:
|
| 155 |
cfg = OmegaConf.load(decoder_config_path)
|
| 156 |
decoder = DVAE(**cfg).to(device=device, dtype=dtype_decoder).eval()
|
| 157 |
assert decoder_ckpt_path, "decoder_ckpt_path should not be None"
|
| 158 |
-
decoder.load_state_dict(
|
|
|
|
|
|
|
| 159 |
self.pretrain_models["decoder"] = decoder
|
| 160 |
self.logger.log(logging.INFO, "decoder loaded.")
|
| 161 |
|
| 162 |
if tokenizer_path:
|
| 163 |
-
tokenizer = torch.load(tokenizer_path, map_location=
|
| 164 |
tokenizer.padding_side = "left"
|
| 165 |
self.pretrain_models["tokenizer"] = tokenizer
|
| 166 |
self.logger.log(logging.INFO, "tokenizer loaded.")
|
|
|
|
| 112 |
dtype_gpt = dtype_gpt or dtype
|
| 113 |
dtype_decoder = dtype_decoder or dtype
|
| 114 |
|
| 115 |
+
map_location = torch.device("cpu")
|
| 116 |
+
|
| 117 |
if vocos_config_path:
|
| 118 |
vocos = (
|
| 119 |
Vocos.from_hparams(vocos_config_path)
|
|
|
|
| 121 |
.eval()
|
| 122 |
)
|
| 123 |
assert vocos_ckpt_path, "vocos_ckpt_path should not be None"
|
| 124 |
+
vocos.load_state_dict(
|
| 125 |
+
torch.load(vocos_ckpt_path, map_location=map_location)
|
| 126 |
+
)
|
| 127 |
self.pretrain_models["vocos"] = vocos
|
| 128 |
self.logger.log(logging.INFO, "vocos loaded.")
|
| 129 |
|
|
|
|
| 131 |
cfg = OmegaConf.load(dvae_config_path)
|
| 132 |
dvae = DVAE(**cfg).to(device=device, dtype=dtype_dvae).eval()
|
| 133 |
assert dvae_ckpt_path, "dvae_ckpt_path should not be None"
|
| 134 |
+
dvae.load_state_dict(torch.load(dvae_ckpt_path, map_location=map_location))
|
| 135 |
self.pretrain_models["dvae"] = dvae
|
| 136 |
self.logger.log(logging.INFO, "dvae loaded.")
|
| 137 |
|
|
|
|
| 139 |
cfg = OmegaConf.load(gpt_config_path)
|
| 140 |
gpt = GPT_warpper(**cfg).to(device=device, dtype=dtype_gpt).eval()
|
| 141 |
assert gpt_ckpt_path, "gpt_ckpt_path should not be None"
|
| 142 |
+
gpt.load_state_dict(torch.load(gpt_ckpt_path, map_location=map_location))
|
| 143 |
if compile and "cuda" in str(device):
|
| 144 |
self.logger.info("compile gpt model")
|
| 145 |
gpt.gpt.forward = torch.compile(
|
|
|
|
| 150 |
assert os.path.exists(
|
| 151 |
spk_stat_path
|
| 152 |
), f"Missing spk_stat.pt: {spk_stat_path}"
|
| 153 |
+
self.pretrain_models["spk_stat"] = torch.load(
|
| 154 |
+
spk_stat_path, map_location=map_location
|
| 155 |
+
).to(device=device, dtype=dtype)
|
| 156 |
self.logger.log(logging.INFO, "gpt loaded.")
|
| 157 |
|
| 158 |
if decoder_config_path:
|
| 159 |
cfg = OmegaConf.load(decoder_config_path)
|
| 160 |
decoder = DVAE(**cfg).to(device=device, dtype=dtype_decoder).eval()
|
| 161 |
assert decoder_ckpt_path, "decoder_ckpt_path should not be None"
|
| 162 |
+
decoder.load_state_dict(
|
| 163 |
+
torch.load(decoder_ckpt_path, map_location=map_location)
|
| 164 |
+
)
|
| 165 |
self.pretrain_models["decoder"] = decoder
|
| 166 |
self.logger.log(logging.INFO, "decoder loaded.")
|
| 167 |
|
| 168 |
if tokenizer_path:
|
| 169 |
+
tokenizer = torch.load(tokenizer_path, map_location=map_location)
|
| 170 |
tokenizer.padding_side = "left"
|
| 171 |
self.pretrain_models["tokenizer"] = tokenizer
|
| 172 |
self.logger.log(logging.INFO, "tokenizer loaded.")
|
modules/Enhancer/ResembleEnhance.py
CHANGED
|
@@ -1,6 +1,5 @@
|
|
| 1 |
import gc
|
| 2 |
-
import
|
| 3 |
-
from typing import List, Literal
|
| 4 |
|
| 5 |
import numpy as np
|
| 6 |
from modules.devices import devices
|
|
@@ -14,7 +13,6 @@ from modules.utils.constants import MODELS_DIR
|
|
| 14 |
from pathlib import Path
|
| 15 |
|
| 16 |
from threading import Lock
|
| 17 |
-
from modules import config
|
| 18 |
|
| 19 |
import logging
|
| 20 |
|
|
@@ -34,12 +32,13 @@ class ResembleEnhance:
|
|
| 34 |
|
| 35 |
def load_model(self):
|
| 36 |
hparams = HParams.load(Path(MODELS_DIR) / "resemble-enhance")
|
| 37 |
-
enhancer = Enhancer(hparams)
|
| 38 |
state_dict = torch.load(
|
| 39 |
Path(MODELS_DIR) / "resemble-enhance" / "mp_rank_00_model_states.pt",
|
| 40 |
-
map_location=
|
| 41 |
)["module"]
|
| 42 |
enhancer.load_state_dict(state_dict)
|
|
|
|
| 43 |
|
| 44 |
self.hparams = hparams
|
| 45 |
self.enhancer = enhancer
|
|
|
|
| 1 |
import gc
|
| 2 |
+
from typing import Literal
|
|
|
|
| 3 |
|
| 4 |
import numpy as np
|
| 5 |
from modules.devices import devices
|
|
|
|
| 13 |
from pathlib import Path
|
| 14 |
|
| 15 |
from threading import Lock
|
|
|
|
| 16 |
|
| 17 |
import logging
|
| 18 |
|
|
|
|
| 32 |
|
| 33 |
def load_model(self):
|
| 34 |
hparams = HParams.load(Path(MODELS_DIR) / "resemble-enhance")
|
| 35 |
+
enhancer = Enhancer(hparams)
|
| 36 |
state_dict = torch.load(
|
| 37 |
Path(MODELS_DIR) / "resemble-enhance" / "mp_rank_00_model_states.pt",
|
| 38 |
+
map_location="cpu",
|
| 39 |
)["module"]
|
| 40 |
enhancer.load_state_dict(state_dict)
|
| 41 |
+
enhancer.to(device=self.device, dtype=self.dtype).eval()
|
| 42 |
|
| 43 |
self.hparams = hparams
|
| 44 |
self.enhancer = enhancer
|
modules/api/api_setup.py
CHANGED
|
@@ -1,9 +1,10 @@
|
|
| 1 |
import logging
|
|
|
|
| 2 |
from modules.devices import devices
|
| 3 |
import argparse
|
| 4 |
|
| 5 |
-
import torch
|
| 6 |
from modules import config
|
|
|
|
| 7 |
from modules.utils import env
|
| 8 |
from modules import generate_audio
|
| 9 |
from modules.api.Api import APIManager
|
|
@@ -77,6 +78,11 @@ def setup_model_args(parser: argparse.ArgumentParser):
|
|
| 77 |
action="store_true",
|
| 78 |
help="Enable debug mode for audio generation",
|
| 79 |
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 80 |
|
| 81 |
|
| 82 |
def process_model_args(args):
|
|
@@ -87,6 +93,7 @@ def process_model_args(args):
|
|
| 87 |
no_half = env.get_and_update_env(args, "no_half", False, bool)
|
| 88 |
off_tqdm = env.get_and_update_env(args, "off_tqdm", False, bool)
|
| 89 |
debug_generate = env.get_and_update_env(args, "debug_generate", False, bool)
|
|
|
|
| 90 |
|
| 91 |
generate_audio.setup_lru_cache()
|
| 92 |
devices.reset_device()
|
|
@@ -95,6 +102,10 @@ def process_model_args(args):
|
|
| 95 |
if debug_generate:
|
| 96 |
generate_audio.logger.setLevel(logging.DEBUG)
|
| 97 |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 98 |
|
| 99 |
def setup_uvicon_args(parser: argparse.ArgumentParser):
|
| 100 |
parser.add_argument("--host", type=str, help="Host to run the server on")
|
|
|
|
| 1 |
import logging
|
| 2 |
+
from modules.Enhancer.ResembleEnhance import load_enhancer
|
| 3 |
from modules.devices import devices
|
| 4 |
import argparse
|
| 5 |
|
|
|
|
| 6 |
from modules import config
|
| 7 |
+
from modules.models import load_chat_tts
|
| 8 |
from modules.utils import env
|
| 9 |
from modules import generate_audio
|
| 10 |
from modules.api.Api import APIManager
|
|
|
|
| 78 |
action="store_true",
|
| 79 |
help="Enable debug mode for audio generation",
|
| 80 |
)
|
| 81 |
+
parser.add_argument(
|
| 82 |
+
"--preload_models",
|
| 83 |
+
action="store_true",
|
| 84 |
+
help="Preload all models at startup",
|
| 85 |
+
)
|
| 86 |
|
| 87 |
|
| 88 |
def process_model_args(args):
|
|
|
|
| 93 |
no_half = env.get_and_update_env(args, "no_half", False, bool)
|
| 94 |
off_tqdm = env.get_and_update_env(args, "off_tqdm", False, bool)
|
| 95 |
debug_generate = env.get_and_update_env(args, "debug_generate", False, bool)
|
| 96 |
+
preload_models = env.get_and_update_env(args, "preload_models", False, bool)
|
| 97 |
|
| 98 |
generate_audio.setup_lru_cache()
|
| 99 |
devices.reset_device()
|
|
|
|
| 102 |
if debug_generate:
|
| 103 |
generate_audio.logger.setLevel(logging.DEBUG)
|
| 104 |
|
| 105 |
+
if preload_models:
|
| 106 |
+
load_chat_tts()
|
| 107 |
+
load_enhancer()
|
| 108 |
+
|
| 109 |
|
| 110 |
def setup_uvicon_args(parser: argparse.ArgumentParser):
|
| 111 |
parser.add_argument("--host", type=str, help="Host to run the server on")
|
modules/webui/css/style.css
CHANGED
|
@@ -66,9 +66,9 @@
|
|
| 66 |
display: none !important;
|
| 67 |
}
|
| 68 |
|
| 69 |
-
.progress-bar{
|
| 70 |
height: 30px !important;
|
| 71 |
-
}
|
| 72 |
|
| 73 |
.progress-bar span {
|
| 74 |
text-align: right;
|
|
|
|
| 66 |
display: none !important;
|
| 67 |
}
|
| 68 |
|
| 69 |
+
/* .progress-bar{
|
| 70 |
height: 30px !important;
|
| 71 |
+
} */
|
| 72 |
|
| 73 |
.progress-bar span {
|
| 74 |
text-align: right;
|
modules/webui/webui_utils.py
CHANGED
|
@@ -1,14 +1,13 @@
|
|
| 1 |
-
import io
|
| 2 |
from typing import Union
|
| 3 |
import numpy as np
|
| 4 |
|
| 5 |
from modules.Enhancer.ResembleEnhance import apply_audio_enhance as _apply_audio_enhance
|
| 6 |
-
from modules.devices import devices
|
| 7 |
from modules.synthesize_audio import synthesize_audio
|
| 8 |
from modules.utils.hf import spaces
|
| 9 |
from modules.webui import webui_config
|
| 10 |
|
| 11 |
import torch
|
|
|
|
| 12 |
|
| 13 |
from modules.ssml_parser.SSMLParser import create_ssml_parser, SSMLBreak, SSMLSegment
|
| 14 |
from modules.SynthesizeSegments import SynthesizeSegments, combine_audio_segments
|
|
@@ -24,7 +23,6 @@ from modules import refiner
|
|
| 24 |
from modules.utils import audio
|
| 25 |
from modules.SentenceSplitter import SentenceSplitter
|
| 26 |
|
| 27 |
-
from pydub import AudioSegment
|
| 28 |
import torch.profiler
|
| 29 |
|
| 30 |
|
|
@@ -97,6 +95,7 @@ def synthesize_ssml(
|
|
| 97 |
enable_denoise=False,
|
| 98 |
eos: str = "[uv_break]",
|
| 99 |
spliter_thr: int = 100,
|
|
|
|
| 100 |
):
|
| 101 |
try:
|
| 102 |
batch_size = int(batch_size)
|
|
@@ -157,6 +156,7 @@ def tts_generate(
|
|
| 157 |
spk_file=None,
|
| 158 |
spliter_thr: int = 100,
|
| 159 |
eos: str = "[uv_break]",
|
|
|
|
| 160 |
):
|
| 161 |
try:
|
| 162 |
batch_size = int(batch_size)
|
|
@@ -219,7 +219,11 @@ def tts_generate(
|
|
| 219 |
|
| 220 |
@torch.inference_mode()
|
| 221 |
@spaces.GPU(duration=120)
|
| 222 |
-
def refine_text(
|
|
|
|
|
|
|
|
|
|
|
|
|
| 223 |
text = text_normalize(text)
|
| 224 |
return refiner.refine_text(text, prompt=prompt)
|
| 225 |
|
|
|
|
|
|
|
| 1 |
from typing import Union
|
| 2 |
import numpy as np
|
| 3 |
|
| 4 |
from modules.Enhancer.ResembleEnhance import apply_audio_enhance as _apply_audio_enhance
|
|
|
|
| 5 |
from modules.synthesize_audio import synthesize_audio
|
| 6 |
from modules.utils.hf import spaces
|
| 7 |
from modules.webui import webui_config
|
| 8 |
|
| 9 |
import torch
|
| 10 |
+
import gradio as gr
|
| 11 |
|
| 12 |
from modules.ssml_parser.SSMLParser import create_ssml_parser, SSMLBreak, SSMLSegment
|
| 13 |
from modules.SynthesizeSegments import SynthesizeSegments, combine_audio_segments
|
|
|
|
| 23 |
from modules.utils import audio
|
| 24 |
from modules.SentenceSplitter import SentenceSplitter
|
| 25 |
|
|
|
|
| 26 |
import torch.profiler
|
| 27 |
|
| 28 |
|
|
|
|
| 95 |
enable_denoise=False,
|
| 96 |
eos: str = "[uv_break]",
|
| 97 |
spliter_thr: int = 100,
|
| 98 |
+
progress=gr.Progress(track_tqdm=True),
|
| 99 |
):
|
| 100 |
try:
|
| 101 |
batch_size = int(batch_size)
|
|
|
|
| 156 |
spk_file=None,
|
| 157 |
spliter_thr: int = 100,
|
| 158 |
eos: str = "[uv_break]",
|
| 159 |
+
progress=gr.Progress(track_tqdm=True),
|
| 160 |
):
|
| 161 |
try:
|
| 162 |
batch_size = int(batch_size)
|
|
|
|
| 219 |
|
| 220 |
@torch.inference_mode()
|
| 221 |
@spaces.GPU(duration=120)
|
| 222 |
+
def refine_text(
|
| 223 |
+
text: str,
|
| 224 |
+
prompt: str,
|
| 225 |
+
progress=gr.Progress(track_tqdm=True),
|
| 226 |
+
):
|
| 227 |
text = text_normalize(text)
|
| 228 |
return refiner.refine_text(text, prompt=prompt)
|
| 229 |
|