Spaces:
Running
on
Zero
Running
on
Zero
File size: 5,655 Bytes
076829a a13411c 076829a a13411c 076829a fadb220 076829a fadb220 f1e374a 9ab298d 53a3c1b fadb220 53a3c1b 076829a a13411c 076829a 02f941d f4e70e2 a13411c 076829a a13411c 076829a 02f941d f4e70e2 a13411c 076829a a13411c 076829a 02f941d f4e70e2 a13411c 076829a a13411c 076829a 02f941d f4e70e2 a13411c |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 |
# Copyright (c) 2024 Alibaba Inc (authors: Xiang Lyu)
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import os
import time
from hyperpyyaml import load_hyperpyyaml
from modelscope import snapshot_download
from cosyvoice.cli.frontend import CosyVoiceFrontEnd
from cosyvoice.cli.model import CosyVoiceModel
from cosyvoice.utils.file_utils import logging
class CosyVoice:
def __init__(self, model_dir, load_jit=True, load_trt=False, load_onnx=True, use_fp16=False):
instruct = True if '-Instruct' in model_dir else False
self.model_dir = model_dir
if not os.path.exists(model_dir):
model_dir = snapshot_download(model_dir)
with open('{}/cosyvoice.yaml'.format(model_dir), 'r') as f:
configs = load_hyperpyyaml(f)
self.frontend = CosyVoiceFrontEnd(configs['get_tokenizer'],
configs['feat_extractor'],
'{}/campplus.onnx'.format(model_dir),
'{}/speech_tokenizer_v1.onnx'.format(model_dir),
'{}/spk2info.pt'.format(model_dir),
instruct,
configs['allowed_special'])
self.model = CosyVoiceModel(configs['llm'], configs['flow'], configs['hift'])
self.model.load('{}/llm.pt'.format(model_dir),
'{}/flow.pt'.format(model_dir),
'{}/hift.pt'.format(model_dir))
if load_jit:
self.model.load_jit('{}/llm.text_encoder.fp16.zip'.format(model_dir),
'{}/llm.llm.fp16.zip'.format(model_dir))
# if load_trt:
# self.model.load_trt(model_dir, use_fp16)
if load_onnx:
self.model.load_onnx(model_dir, use_fp16)
del configs
def list_avaliable_spks(self):
spks = list(self.frontend.spk2info.keys())
return spks
def inference_sft(self, tts_text, spk_id, stream=False):
for i in self.frontend.text_normalize(tts_text, split=True):
model_input = self.frontend.frontend_sft(i, spk_id)
start_time = time.time()
logging.info('synthesis text {}'.format(i))
for model_output in self.model.inference(**model_input, stream=stream):
speech_len = model_output['tts_speech'].shape[1] / 22050
logging.info('yield speech len {}, rtf {}'.format(speech_len, (time.time() - start_time) / speech_len))
yield model_output
start_time = time.time()
def inference_zero_shot(self, tts_text, prompt_text, prompt_speech_16k, stream=False):
prompt_text = self.frontend.text_normalize(prompt_text, split=False)
for i in self.frontend.text_normalize(tts_text, split=True):
model_input = self.frontend.frontend_zero_shot(i, prompt_text, prompt_speech_16k)
start_time = time.time()
logging.info('synthesis text {}'.format(i))
for model_output in self.model.inference(**model_input, stream=stream):
speech_len = model_output['tts_speech'].shape[1] / 22050
logging.info('yield speech len {}, rtf {}'.format(speech_len, (time.time() - start_time) / speech_len))
yield model_output
start_time = time.time()
def inference_cross_lingual(self, tts_text, prompt_speech_16k, stream=False):
if self.frontend.instruct is True:
raise ValueError('{} do not support cross_lingual inference'.format(self.model_dir))
for i in self.frontend.text_normalize(tts_text, split=True):
model_input = self.frontend.frontend_cross_lingual(i, prompt_speech_16k)
start_time = time.time()
logging.info('synthesis text {}'.format(i))
for model_output in self.model.inference(**model_input, stream=stream):
speech_len = model_output['tts_speech'].shape[1] / 22050
logging.info('yield speech len {}, rtf {}'.format(speech_len, (time.time() - start_time) / speech_len))
yield model_output
start_time = time.time()
def inference_instruct(self, tts_text, spk_id, instruct_text, stream=False):
if self.frontend.instruct is False:
raise ValueError('{} do not support instruct inference'.format(self.model_dir))
instruct_text = self.frontend.text_normalize(instruct_text, split=False)
for i in self.frontend.text_normalize(tts_text, split=True):
model_input = self.frontend.frontend_instruct(i, spk_id, instruct_text)
start_time = time.time()
logging.info('synthesis text {}'.format(i))
for model_output in self.model.inference(**model_input, stream=stream):
speech_len = model_output['tts_speech'].shape[1] / 22050
logging.info('yield speech len {}, rtf {}'.format(speech_len, (time.time() - start_time) / speech_len))
yield model_output
start_time = time.time()
|