Originals/ViGent2
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases 53 Wiki Activity

更新

Browse Source
This commit is contained in:
Kevin Wong
2026-02-11 13:48:45 +08:00
parent e33dfc3031
commit 96a298e51c
282 changed files with 93514 additions and 461 deletions
Download Patch File Download Diff File Expand all files Collapse all files

13
models/CosyVoice/runtime/python/Dockerfile Normal file
View File

@@ -0,0 +1,13 @@
FROM pytorch/pytorch:2.0.1-cuda11.7-cudnn8-runtime
ENV DEBIAN_FRONTEND=noninteractive
WORKDIR /opt/CosyVoice
RUN sed -i s@/archive.ubuntu.com/@/mirrors.aliyun.com/@g /etc/apt/sources.list
RUN apt-get update -y
RUN apt-get -y install git unzip git-lfs g++
RUN git lfs install
RUN git clone --recursive https://github.com/FunAudioLLM/CosyVoice.git
# here we use python==3.10 because we cannot find an image which have both python3.8 and torch2.0.1-cu118 installed
RUN cd CosyVoice && pip3 install -r requirements.txt -i https://mirrors.aliyun.com/pypi/simple/ --trusted-host=mirrors.aliyun.com --no-cache-dir
RUN cd CosyVoice/runtime/python/grpc && python3 -m grpc_tools.protoc -I. --python_out=. --grpc_python_out=. cosyvoice.proto

92
models/CosyVoice/runtime/python/fastapi/client.py Normal file
View File

@@ -0,0 +1,92 @@
# 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 argparse
import logging
import requests
import torch
import torchaudio
import numpy as np
def main():
url = "http://{}:{}/inference_{}".format(args.host, args.port, args.mode)
if args.mode == 'sft':
payload = {
'tts_text': args.tts_text,
'spk_id': args.spk_id
}
response = requests.request("GET", url, data=payload, stream=True)
elif args.mode == 'zero_shot':
payload = {
'tts_text': args.tts_text,
'prompt_text': args.prompt_text
}
files = [('prompt_wav', ('prompt_wav', open(args.prompt_wav, 'rb'), 'application/octet-stream'))]
response = requests.request("GET", url, data=payload, files=files, stream=True)
elif args.mode == 'cross_lingual':
payload = {
'tts_text': args.tts_text,
}
files = [('prompt_wav', ('prompt_wav', open(args.prompt_wav, 'rb'), 'application/octet-stream'))]
response = requests.request("GET", url, data=payload, files=files, stream=True)
else:
payload = {
'tts_text': args.tts_text,
'spk_id': args.spk_id,
'instruct_text': args.instruct_text
}
response = requests.request("GET", url, data=payload, stream=True)
tts_audio = b''
for r in response.iter_content(chunk_size=16000):
tts_audio += r
tts_speech = torch.from_numpy(np.array(np.frombuffer(tts_audio, dtype=np.int16))).unsqueeze(dim=0)
logging.info('save response to {}'.format(args.tts_wav))
torchaudio.save(args.tts_wav, tts_speech, target_sr)
logging.info('get response')
if __name__ == "__main__":
parser = argparse.ArgumentParser()
parser.add_argument('--host',
type=str,
default='0.0.0.0')
parser.add_argument('--port',
type=int,
default='50000')
parser.add_argument('--mode',
default='sft',
choices=['sft', 'zero_shot', 'cross_lingual', 'instruct'],
help='request mode')
parser.add_argument('--tts_text',
type=str,
default='你好，我是通义千问语音合成大模型，请问有什么可以帮您的吗？')
parser.add_argument('--spk_id',
type=str,
default='中文女')
parser.add_argument('--prompt_text',
type=str,
default='希望你以后能够做的比我还好呦。')
parser.add_argument('--prompt_wav',
type=str,
default='../../../asset/zero_shot_prompt.wav')
parser.add_argument('--instruct_text',
type=str,
default='Theo \'Crimson\', is a fiery, passionate rebel leader. \
Fights with fervor for justice, but struggles with impulsiveness.')
parser.add_argument('--tts_wav',
type=str,
default='demo.wav')
args = parser.parse_args()
prompt_sr, target_sr = 16000, 22050
main()

95
models/CosyVoice/runtime/python/fastapi/server.py Normal file
View File

@@ -0,0 +1,95 @@
# 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 sys
import argparse
import logging
logging.getLogger('matplotlib').setLevel(logging.WARNING)
from fastapi import FastAPI, UploadFile, Form, File
from fastapi.responses import StreamingResponse
from fastapi.middleware.cors import CORSMiddleware
import uvicorn
import numpy as np
ROOT_DIR = os.path.dirname(os.path.abspath(__file__))
sys.path.append('{}/../../..'.format(ROOT_DIR))
sys.path.append('{}/../../../third_party/Matcha-TTS'.format(ROOT_DIR))
from cosyvoice.cli.cosyvoice import AutoModel
from cosyvoice.utils.file_utils import load_wav
app = FastAPI()
# set cross region allowance
app.add_middleware(
CORSMiddleware,
allow_origins=["*"],
allow_credentials=True,
allow_methods=["*"],
allow_headers=["*"])
def generate_data(model_output):
for i in model_output:
tts_audio = (i['tts_speech'].numpy() * (2 ** 15)).astype(np.int16).tobytes()
yield tts_audio
@app.get("/inference_sft")
@app.post("/inference_sft")
async def inference_sft(tts_text: str = Form(), spk_id: str = Form()):
model_output = cosyvoice.inference_sft(tts_text, spk_id)
return StreamingResponse(generate_data(model_output))
@app.get("/inference_zero_shot")
@app.post("/inference_zero_shot")
async def inference_zero_shot(tts_text: str = Form(), prompt_text: str = Form(), prompt_wav: UploadFile = File()):
prompt_speech_16k = load_wav(prompt_wav.file, 16000)
model_output = cosyvoice.inference_zero_shot(tts_text, prompt_text, prompt_speech_16k)
return StreamingResponse(generate_data(model_output))
@app.get("/inference_cross_lingual")
@app.post("/inference_cross_lingual")
async def inference_cross_lingual(tts_text: str = Form(), prompt_wav: UploadFile = File()):
prompt_speech_16k = load_wav(prompt_wav.file, 16000)
model_output = cosyvoice.inference_cross_lingual(tts_text, prompt_speech_16k)
return StreamingResponse(generate_data(model_output))
@app.get("/inference_instruct")
@app.post("/inference_instruct")
async def inference_instruct(tts_text: str = Form(), spk_id: str = Form(), instruct_text: str = Form()):
model_output = cosyvoice.inference_instruct(tts_text, spk_id, instruct_text)
return StreamingResponse(generate_data(model_output))
@app.get("/inference_instruct2")
@app.post("/inference_instruct2")
async def inference_instruct2(tts_text: str = Form(), instruct_text: str = Form(), prompt_wav: UploadFile = File()):
prompt_speech_16k = load_wav(prompt_wav.file, 16000)
model_output = cosyvoice.inference_instruct2(tts_text, instruct_text, prompt_speech_16k)
return StreamingResponse(generate_data(model_output))
if __name__ == '__main__':
parser = argparse.ArgumentParser()
parser.add_argument('--port',
type=int,
default=50000)
parser.add_argument('--model_dir',
type=str,
default='iic/CosyVoice2-0.5B',
help='local path or modelscope repo id')
args = parser.parse_args()
cosyvoice = AutoModel(model_dir=args.model_dir)
uvicorn.run(app, host="0.0.0.0", port=args.port)

106
models/CosyVoice/runtime/python/grpc/client.py Normal file
View File

@@ -0,0 +1,106 @@
# 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 sys
ROOT_DIR = os.path.dirname(os.path.abspath(__file__))
sys.path.append('{}/../../..'.format(ROOT_DIR))
sys.path.append('{}/../../../third_party/Matcha-TTS'.format(ROOT_DIR))
import logging
import argparse
import torchaudio
import cosyvoice_pb2
import cosyvoice_pb2_grpc
import grpc
import torch
import numpy as np
from cosyvoice.utils.file_utils import load_wav
def main():
with grpc.insecure_channel("{}:{}".format(args.host, args.port)) as channel:
stub = cosyvoice_pb2_grpc.CosyVoiceStub(channel)
request = cosyvoice_pb2.Request()
if args.mode == 'sft':
logging.info('send sft request')
sft_request = cosyvoice_pb2.sftRequest()
sft_request.spk_id = args.spk_id
sft_request.tts_text = args.tts_text
request.sft_request.CopyFrom(sft_request)
elif args.mode == 'zero_shot':
logging.info('send zero_shot request')
zero_shot_request = cosyvoice_pb2.zeroshotRequest()
zero_shot_request.tts_text = args.tts_text
zero_shot_request.prompt_text = args.prompt_text
prompt_speech = load_wav(args.prompt_wav, 16000)
zero_shot_request.prompt_audio = (prompt_speech.numpy() * (2**15)).astype(np.int16).tobytes()
request.zero_shot_request.CopyFrom(zero_shot_request)
elif args.mode == 'cross_lingual':
logging.info('send cross_lingual request')
cross_lingual_request = cosyvoice_pb2.crosslingualRequest()
cross_lingual_request.tts_text = args.tts_text
prompt_speech = load_wav(args.prompt_wav, 16000)
cross_lingual_request.prompt_audio = (prompt_speech.numpy() * (2**15)).astype(np.int16).tobytes()
request.cross_lingual_request.CopyFrom(cross_lingual_request)
else:
logging.info('send instruct request')
instruct_request = cosyvoice_pb2.instructRequest()
instruct_request.tts_text = args.tts_text
instruct_request.spk_id = args.spk_id
instruct_request.instruct_text = args.instruct_text
request.instruct_request.CopyFrom(instruct_request)
response = stub.Inference(request)
tts_audio = b''
for r in response:
tts_audio += r.tts_audio
tts_speech = torch.from_numpy(np.array(np.frombuffer(tts_audio, dtype=np.int16))).unsqueeze(dim=0)
logging.info('save response to {}'.format(args.tts_wav))
torchaudio.save(args.tts_wav, tts_speech, target_sr)
logging.info('get response')
if __name__ == "__main__":
parser = argparse.ArgumentParser()
parser.add_argument('--host',
type=str,
default='0.0.0.0')
parser.add_argument('--port',
type=int,
default='50000')
parser.add_argument('--mode',
default='sft',
choices=['sft', 'zero_shot', 'cross_lingual', 'instruct'],
help='request mode')
parser.add_argument('--tts_text',
type=str,
default='你好，我是通义千问语音合成大模型，请问有什么可以帮您的吗？')
parser.add_argument('--spk_id',
type=str,
default='中文女')
parser.add_argument('--prompt_text',
type=str,
default='希望你以后能够做的比我还好呦。')
parser.add_argument('--prompt_wav',
type=str,
default='../../../asset/zero_shot_prompt.wav')
parser.add_argument('--instruct_text',
type=str,
default='Theo \'Crimson\', is a fiery, passionate rebel leader. \
Fights with fervor for justice, but struggles with impulsiveness.')
parser.add_argument('--tts_wav',
type=str,
default='demo.wav')
args = parser.parse_args()
prompt_sr, target_sr = 16000, 22050
main()

43
models/CosyVoice/runtime/python/grpc/cosyvoice.proto Normal file
View File

@@ -0,0 +1,43 @@
syntax = "proto3";
package cosyvoice;
option go_package = "protos/";
service CosyVoice{
rpc Inference(Request) returns (stream Response) {}
}
message Request{
oneof RequestPayload {
sftRequest sft_request = 1;
zeroshotRequest zero_shot_request = 2;
crosslingualRequest cross_lingual_request = 3;
instructRequest instruct_request = 4;
}
}
message sftRequest{
string spk_id = 1;
string tts_text = 2;
}
message zeroshotRequest{
string tts_text = 1;
string prompt_text = 2;
bytes prompt_audio = 3;
}
message crosslingualRequest{
string tts_text = 1;
bytes prompt_audio = 2;
}
message instructRequest{
string tts_text = 1;
string spk_id = 2;
string instruct_text = 3;
}
message Response{
bytes tts_audio = 1;
}

90
models/CosyVoice/runtime/python/grpc/server.py Normal file
View File

@@ -0,0 +1,90 @@
# 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 sys
from concurrent import futures
import argparse
import cosyvoice_pb2
import cosyvoice_pb2_grpc
import logging
logging.getLogger('matplotlib').setLevel(logging.WARNING)
import grpc
import torch
import numpy as np
ROOT_DIR = os.path.dirname(os.path.abspath(__file__))
sys.path.append('{}/../../..'.format(ROOT_DIR))
sys.path.append('{}/../../../third_party/Matcha-TTS'.format(ROOT_DIR))
from cosyvoice.cli.cosyvoice import AutoModel
logging.basicConfig(level=logging.DEBUG,
format='%(asctime)s %(levelname)s %(message)s')
class CosyVoiceServiceImpl(cosyvoice_pb2_grpc.CosyVoiceServicer):
def __init__(self, args):
self.cosyvoice = AutoModel(model_dir=args.model_dir)
logging.info('grpc service initialized')
def Inference(self, request, context):
if request.HasField('sft_request'):
logging.info('get sft inference request')
model_output = self.cosyvoice.inference_sft(request.sft_request.tts_text, request.sft_request.spk_id)
elif request.HasField('zero_shot_request'):
logging.info('get zero_shot inference request')
prompt_speech_16k = torch.from_numpy(np.array(np.frombuffer(request.zero_shot_request.prompt_audio, dtype=np.int16))).unsqueeze(dim=0)
prompt_speech_16k = prompt_speech_16k.float() / (2**15)
model_output = self.cosyvoice.inference_zero_shot(request.zero_shot_request.tts_text,
request.zero_shot_request.prompt_text,
prompt_speech_16k)
elif request.HasField('cross_lingual_request'):
logging.info('get cross_lingual inference request')
prompt_speech_16k = torch.from_numpy(np.array(np.frombuffer(request.cross_lingual_request.prompt_audio, dtype=np.int16))).unsqueeze(dim=0)
prompt_speech_16k = prompt_speech_16k.float() / (2**15)
model_output = self.cosyvoice.inference_cross_lingual(request.cross_lingual_request.tts_text, prompt_speech_16k)
else:
logging.info('get instruct inference request')
model_output = self.cosyvoice.inference_instruct(request.instruct_request.tts_text,
request.instruct_request.spk_id,
request.instruct_request.instruct_text)
logging.info('send inference response')
for i in model_output:
response = cosyvoice_pb2.Response()
response.tts_audio = (i['tts_speech'].numpy() * (2 ** 15)).astype(np.int16).tobytes()
yield response
def main():
grpcServer = grpc.server(futures.ThreadPoolExecutor(max_workers=args.max_conc), maximum_concurrent_rpcs=args.max_conc)
cosyvoice_pb2_grpc.add_CosyVoiceServicer_to_server(CosyVoiceServiceImpl(args), grpcServer)
grpcServer.add_insecure_port('0.0.0.0:{}'.format(args.port))
grpcServer.start()
logging.info("server listening on 0.0.0.0:{}".format(args.port))
grpcServer.wait_for_termination()
if __name__ == '__main__':
parser = argparse.ArgumentParser()
parser.add_argument('--port',
type=int,
default=50000)
parser.add_argument('--max_conc',
type=int,
default=4)
parser.add_argument('--model_dir',
type=str,
default='iic/CosyVoice2-0.5B',
help='local path or modelscope repo id')
args = parser.parse_args()
main()