封装minmax语音克隆接口
This commit is contained in:
gexianmeng
parent
a4c4ee1f5b
commit
3f49577626
|
|
@ -0,0 +1,18 @@
|
|||
import modal
|
||||
from src.cluster.config import config
|
||||
from src.cluster.video_downloader.worker import worker_app
|
||||
from src.cluster.web.worker import fastapi_app
|
||||
from src.cluster.ffmpeg_worker.worker import app as ffmpeg_app
|
||||
from src.cluster.comfyui.worker import app as comfyui_app
|
||||
from src.cluster.comfyui_latentsync15.worker import app as comfyui_latentsync15_app
|
||||
|
||||
app = modal.App(config.app_name,
|
||||
include_source=False,
|
||||
secrets=[modal.Secret.from_name("cf-kv-secret",
|
||||
environment_name=config.environment)])
|
||||
|
||||
app.include(fastapi_app)
|
||||
app.include(worker_app)
|
||||
app.include(ffmpeg_app)
|
||||
app.include(comfyui_app)
|
||||
app.include(comfyui_latentsync15_app)
|
||||
|
|
@ -0,0 +1,321 @@
|
|||
import asyncio
|
||||
import os
|
||||
from pathlib import Path
|
||||
from typing import Optional
|
||||
|
||||
import sentry_sdk
|
||||
from fastapi import APIRouter, Depends, HTTPException, File, UploadFile, Form
|
||||
from fastapi.responses import JSONResponse
|
||||
from loguru import logger
|
||||
from modal import current_function_call_id
|
||||
from pydantic import BaseModel, Field
|
||||
from starlette import status
|
||||
|
||||
from ..middleware.authorization import verify_token
|
||||
from ..models.media_model import MediaSource, MediaCacheStatus
|
||||
from ..utils.KVCache import KVCache
|
||||
from ..utils.SentryUtils import SentryUtils
|
||||
# 导入语音克隆函数
|
||||
from cluster.minmax_clone_audio_api import voice_clone_and_download
|
||||
|
||||
|
||||
# 创建路由，使用 /text2speech 作为前缀
|
||||
router = APIRouter(prefix="/text2speech", tags=["文本转语音"])
|
||||
|
||||
# 使用函数来延迟获取配置
|
||||
def get_config():
|
||||
"""延迟加载配置，确保环境变量已经被 Modal 注入"""
|
||||
from ..config import WorkerConfig
|
||||
return WorkerConfig()
|
||||
|
||||
|
||||
# 定义请求和响应模型
|
||||
class Text2SpeechRequest(BaseModel):
|
||||
"""文本转语音请求模型"""
|
||||
text: str = Field(description="要合成的文本内容")
|
||||
voice_id: Optional[str] = Field(default=None, description="声音ID，如果为None则自动生成")
|
||||
model: str = Field(default="speech-02-hd", description="使用的模型")
|
||||
accuracy: float = Field(default=0.8, description="精确度，范围0-1")
|
||||
|
||||
|
||||
class Text2SpeechResponse(BaseModel):
|
||||
"""文本转语音响应模型"""
|
||||
success: bool = Field(description="是否成功")
|
||||
media_source: Optional[MediaSource] = Field(default=None, description="生成的音频文件信息")
|
||||
download_url: Optional[str] = Field(default=None, description="下载链接")
|
||||
voice_id: Optional[str] = Field(default=None, description="声音ID")
|
||||
error: Optional[str] = Field(default=None, description="错误信息")
|
||||
|
||||
|
||||
@router.post("/clone", summary="语音克隆合成", description="上传音频文件进行语音克隆，并使用克隆的声音合成新的文本音频",
|
||||
dependencies=[Depends(verify_token)])
|
||||
async def voice_clone(file: UploadFile = File(description="上传的音频文件，用于克隆声音"),
|
||||
text: str = Form(description="要合成的文本内容"),
|
||||
model: str = Form(default="speech-02-hd", description="使用的模型"),
|
||||
accuracy: float = Form(default=0.8, description="精确度")) -> Text2SpeechResponse:
|
||||
"""
|
||||
语音克隆接口
|
||||
|
||||
该接口接收一个音频文件，克隆其声音特征，并使用克隆的声音合成新的文本音频。
|
||||
"""
|
||||
fn_id = current_function_call_id()
|
||||
|
||||
# 在函数内部获取配置
|
||||
config = get_config()
|
||||
modal_kv_cache = KVCache(kv_name=config.modal_kv_name, environment=config.modal_environment)
|
||||
|
||||
# 添加配置验证日志
|
||||
logger.info(f"开始语音克隆处理 - Function ID: {fn_id}")
|
||||
logger.info(f"MiniMax Group ID 配置状态: {'已配置' if config.minimax_group_id else '未配置'}")
|
||||
logger.info(f"MiniMax API Key 配置状态: {'已配置' if config.minimax_api_key else '未配置'}")
|
||||
|
||||
# 检查必要的配置
|
||||
if not config.minimax_group_id or not config.minimax_api_key:
|
||||
error_msg = []
|
||||
if not config.minimax_group_id:
|
||||
error_msg.append("MINIMAX_GROUP_ID 未配置")
|
||||
if not config.minimax_api_key:
|
||||
error_msg.append("MINIMAX_API_KEY 未配置")
|
||||
|
||||
logger.error(f"MiniMax API 配置缺失: {', '.join(error_msg)}")
|
||||
return Text2SpeechResponse(success=False, error=f"MiniMax API 配置错误: {', '.join(error_msg)}")
|
||||
|
||||
# 验证文件类型
|
||||
allowed_extensions = {'.mp3', '.wav', '.m4a', '.aac', '.ogg'}
|
||||
file_extension = Path(file.filename).suffix.lower()
|
||||
|
||||
if file_extension not in allowed_extensions:
|
||||
|
||||
logger.warning(f"不支持的文件格式: {file_extension}")
|
||||
raise HTTPException(status_code=status.HTTP_400_BAD_REQUEST,
|
||||
detail=f"不支持的文件格式。支持的格式: {', '.join(allowed_extensions)}")
|
||||
|
||||
# 文件大小限制
|
||||
max_file_size = config.voice_clone_max_file_size
|
||||
if file.size > max_file_size:
|
||||
logger.warning(f"文件大小超限: {file.size} > {max_file_size}")
|
||||
raise HTTPException(status_code=status.HTTP_413_REQUEST_ENTITY_TOO_LARGE,
|
||||
detail=f"文件大小超过限制。最大允许: {max_file_size / 1024 / 1024}MB")
|
||||
|
||||
temp_dir = None
|
||||
temp_audio_path = None
|
||||
|
||||
try:
|
||||
# 创建临时目录保存上传的文件
|
||||
temp_dir = f"/tmp/text2speech_{fn_id}"
|
||||
os.makedirs(temp_dir, exist_ok=True)
|
||||
|
||||
# 保存上传的文件
|
||||
temp_audio_path = os.path.join(temp_dir, file.filename)
|
||||
with open(temp_audio_path, 'wb') as f:
|
||||
content = await file.read()
|
||||
f.write(content)
|
||||
|
||||
logger.info(f"音频文件已保存到: {temp_audio_path}, 大小: {len(content)} bytes")
|
||||
|
||||
# 设置输出目录
|
||||
output_dir = os.path.join(config.S3_mount_dir, config.voice_clone_output_dir)
|
||||
os.makedirs(output_dir, exist_ok=True)
|
||||
logger.info(f"输出目录: {output_dir}")
|
||||
|
||||
# 调用语音克隆函数
|
||||
@SentryUtils.sentry_tracker(name="文本转语音处理", op="text2speech.clone", fn_id=fn_id, sentry_trace_id=None,
|
||||
sentry_baggage=None)
|
||||
async def process_voice_clone():
|
||||
|
||||
try:
|
||||
# 在异步环境中运行同步函数
|
||||
loop = asyncio.get_event_loop()
|
||||
logger.info(f"开始调用 MiniMax API - text: {text[:50]}...")
|
||||
|
||||
result = await loop.run_in_executor(None,
|
||||
voice_clone_and_download,
|
||||
temp_audio_path,
|
||||
config.minimax_group_id,
|
||||
config.minimax_api_key,
|
||||
text,
|
||||
output_dir,
|
||||
model,
|
||||
accuracy)
|
||||
|
||||
logger.info(f"MiniMax API 调用完成 - 结果: {result}")
|
||||
return result
|
||||
|
||||
except Exception as api_error:
|
||||
logger.error(f"MiniMax API 调用异常: {str(api_error)}")
|
||||
raise
|
||||
|
||||
# 执行语音克隆
|
||||
output_path, used_voice_id = await process_voice_clone()
|
||||
|
||||
logger.info(f"语音克隆结果 - output_path: {output_path}, voice_id: {used_voice_id}")
|
||||
|
||||
# 检查结果
|
||||
if output_path is None:
|
||||
logger.error("语音合成失败: output_path 为 None")
|
||||
return Text2SpeechResponse(
|
||||
success=False,
|
||||
voice_id=used_voice_id or "",
|
||||
error="语音合成失败：MiniMax API 返回空结果，请检查 API 配置和余额"
|
||||
)
|
||||
|
||||
# 验证输出文件是否存在
|
||||
if not os.path.exists(output_path):
|
||||
logger.error(f"输出文件不存在: {output_path}")
|
||||
return Text2SpeechResponse(
|
||||
success=False,
|
||||
voice_id=used_voice_id or "",
|
||||
error=f"生成的音频文件不存在: {output_path}"
|
||||
)
|
||||
|
||||
# 创建 MediaSource 对象
|
||||
# 计算相对于S3挂载目录的路径
|
||||
relative_path = os.path.relpath(output_path, config.S3_mount_dir)
|
||||
s3_key = relative_path.replace("\\", "/") # Windows兼容
|
||||
|
||||
media_source = MediaSource(
|
||||
path=s3_key,
|
||||
protocol="s3",
|
||||
endpoint=config.S3_region,
|
||||
bucket=config.S3_bucket_name,
|
||||
urn=f"s3://{config.S3_region}/{config.S3_bucket_name}/{s3_key}",
|
||||
status=MediaCacheStatus.ready, downloader_id=fn_id
|
||||
)
|
||||
|
||||
# 保存到KV缓存
|
||||
modal_kv_cache.set_cache(media_source)
|
||||
logger.info(f"已保存到 KV 缓存: {media_source.urn}")
|
||||
|
||||
# 生成下载链接
|
||||
download_url = f"{config.S3_cdn_endpoint}/{s3_key}"
|
||||
|
||||
logger.success(f"文本转语音成功完成。Voice ID: {used_voice_id}, Download URL: {download_url}")
|
||||
|
||||
return Text2SpeechResponse(
|
||||
success=True,
|
||||
voice_id=used_voice_id,
|
||||
media_source=media_source,
|
||||
download_url=download_url
|
||||
)
|
||||
|
||||
except HTTPException:
|
||||
# 重新抛出 HTTP 异常
|
||||
raise
|
||||
except Exception as e:
|
||||
logger.error(f"文本转语音处理失败 - 详细错误: {str(e)}", exc_info=True)
|
||||
sentry_sdk.capture_exception(e)
|
||||
|
||||
# 返回详细的错误信息
|
||||
error_detail = str(e)
|
||||
if "401" in error_detail or "403" in error_detail:
|
||||
error_msg = "MiniMax API 认证失败，请检查 API Key"
|
||||
elif "429" in error_detail:
|
||||
error_msg = "MiniMax API 请求过于频繁，请稍后重试"
|
||||
elif "insufficient" in error_detail.lower() or "quota" in error_detail.lower():
|
||||
error_msg = "MiniMax API 余额不足"
|
||||
else:
|
||||
error_msg = f"处理失败: {error_detail[:200]}" # 限制错误信息长度
|
||||
|
||||
return Text2SpeechResponse(success=False, error=error_msg)
|
||||
|
||||
finally:
|
||||
# 清理临时文件
|
||||
try:
|
||||
if temp_audio_path and os.path.exists(temp_audio_path):
|
||||
os.remove(temp_audio_path)
|
||||
logger.info(f"已删除临时文件: {temp_audio_path}")
|
||||
if temp_dir and os.path.exists(temp_dir):
|
||||
os.rmdir(temp_dir)
|
||||
logger.info(f"已删除临时目录: {temp_dir}")
|
||||
|
||||
except Exception as cleanup_error:
|
||||
logger.warning(f"清理临时文件失败: {cleanup_error}")
|
||||
|
||||
|
||||
@router.get("/voice/{voice_id}", summary="查询声音状态", description="根据voice_id查询声音文件的状态")
|
||||
async def get_voice_status(voice_id: str):
|
||||
"""
|
||||
查询特定voice_id的处理状态
|
||||
"""
|
||||
# 在函数内部获取配置
|
||||
config = get_config()
|
||||
modal_kv_cache = KVCache(kv_name=config.modal_kv_name, environment=config.modal_environment)
|
||||
|
||||
try:
|
||||
# 尝试从KV缓存中查找相关的媒体文件
|
||||
key_pattern = f"{config.voice_clone_output_dir}/{voice_id}.mp3"
|
||||
urn = f"s3://{config.S3_region}/{config.S3_bucket_name}/{key_pattern}"
|
||||
|
||||
cached_media = modal_kv_cache.get_cache(urn)
|
||||
|
||||
if cached_media:
|
||||
download_url = f"{config.S3_cdn_endpoint}/{cached_media.path}"
|
||||
return JSONResponse(content={
|
||||
"success": True,
|
||||
"voice_id": voice_id,
|
||||
"status": cached_media.status.value,
|
||||
"download_url": download_url if cached_media.status == MediaCacheStatus.ready else None,
|
||||
"media_source": cached_media.model_dump()
|
||||
})
|
||||
|
||||
else:
|
||||
return JSONResponse(status_code=status.HTTP_404_NOT_FOUND,
|
||||
content={"success": False, "voice_id": voice_id, "error": "未找到该voice_id的记录"})
|
||||
except Exception as e:
|
||||
logger.error(f"查询声音状态失败: {str(e)}")
|
||||
return JSONResponse(
|
||||
status_code=status.HTTP_500_INTERNAL_SERVER_ERROR,
|
||||
content={"success": False, "voice_id": voice_id, "error": f"查询失败: {str(e)}"}
|
||||
)
|
||||
|
||||
|
||||
# 调试端点
|
||||
@router.get("/debug/env",
|
||||
summary="调试环境变量",
|
||||
description="检查环境变量和配置状态",
|
||||
dependencies=[Depends(verify_token)])
|
||||
async def debug_environment():
|
||||
"""调试端点，检查环境变量是否正确加载"""
|
||||
import os
|
||||
|
||||
# 获取所有环境变量
|
||||
all_env_vars = dict(os.environ)
|
||||
|
||||
# 查找 MINIMAX 相关的环境变量
|
||||
minimax_vars = {k: v for k, v in all_env_vars.items() if 'MINIMAX' in k.upper()}
|
||||
|
||||
# 尝试加载配置
|
||||
try:
|
||||
config = get_config()
|
||||
config_loaded = True
|
||||
config_values = {
|
||||
"minimax_group_id": bool(config.minimax_group_id),
|
||||
"minimax_api_key": bool(config.minimax_api_key),
|
||||
"minimax_group_id_length": len(config.minimax_group_id),
|
||||
"minimax_api_key_length": len(config.minimax_api_key),
|
||||
}
|
||||
except Exception as e:
|
||||
config_loaded = False
|
||||
config_values = {"error": str(e)}
|
||||
|
||||
return {
|
||||
"environment": {
|
||||
"MODAL_ENVIRONMENT": os.getenv("MODAL_ENVIRONMENT", "NOT_SET"),
|
||||
"total_env_vars": len(all_env_vars),
|
||||
"minimax_env_vars_found": len(minimax_vars),
|
||||
},
|
||||
"minimax_env_vars": {
|
||||
k: f"{v[:20]}..." if len(v) > 20 else v
|
||||
for k, v in minimax_vars.items()
|
||||
},
|
||||
"direct_env_check": {
|
||||
"MINIMAX_GROUP_ID": os.getenv("MINIMAX_GROUP_ID", "NOT_FOUND"),
|
||||
"MINIMAX_API_KEY": "FOUND" if os.getenv("MINIMAX_API_KEY") else "NOT_FOUND",
|
||||
},
|
||||
"config_loaded": config_loaded,
|
||||
"config_values": config_values,
|
||||
"secrets_in_env": {
|
||||
"CF_KV_SECRET": "cf-kv" in str(all_env_vars),
|
||||
"MINIMAX_SECRET": any("minimax" in k.lower() for k in all_env_vars.keys()),
|
||||
}
|
||||
}
|
||||
|
|
@ -0,0 +1,146 @@
|
|||
import json
|
||||
import os
|
||||
import uuid
|
||||
from datetime import datetime
|
||||
|
||||
import requests
|
||||
from loguru import logger
|
||||
|
||||
|
||||
def generate_unique_voice_id(prefix="minmax_audio_clone"):
|
||||
"""
|
||||
生成唯一的voice_id
|
||||
|
||||
参数:
|
||||
prefix: voice_id的前缀，默认为"minmax_audio_clone"
|
||||
|
||||
返回:
|
||||
符合MiniMax规则的唯一voice_id，格式：minmax_audio_clone_20250803_547218f2
|
||||
"""
|
||||
# 获取当前日期，格式：YYYYMMDD
|
||||
current_date = datetime.now().strftime("%Y%m%d")
|
||||
|
||||
# 生成8位随机字符
|
||||
unique_id = str(uuid.uuid4()).replace('-', '')[:8]
|
||||
|
||||
# 组合voice_id
|
||||
voice_id = f"{prefix}_{current_date}_{unique_id}"
|
||||
|
||||
# 确保长度不超过256个字符
|
||||
if len(voice_id) > 256:
|
||||
voice_id = voice_id[:256]
|
||||
|
||||
return voice_id
|
||||
|
||||
|
||||
def voice_clone_and_download(audio_file_path, group_id, api_key, text, output_dir, model="speech-02-hd",
|
||||
accuracy=0.8):
|
||||
"""
|
||||
音频复刻并下载生成的音频文件
|
||||
|
||||
参数:
|
||||
audio_file_path: 输入音频文件路径
|
||||
group_id: MiniMax API的组ID
|
||||
api_key: MiniMax API密钥
|
||||
text: 要合成的文本内容
|
||||
output_dir: 输出目录
|
||||
model: 使用的模型，默认为"speech-02-hd"
|
||||
accuracy: 精确度，默认为0.8
|
||||
|
||||
返回:
|
||||
(下载的音频文件完整路径, 使用的voice_id)的元组，如果失败返回(None, None)
|
||||
"""
|
||||
|
||||
try:
|
||||
# 如果没有提供voice_id，则生成一个唯一的
|
||||
voice_id = generate_unique_voice_id()
|
||||
|
||||
logger.info(f"使用voice_id: {voice_id}")
|
||||
|
||||
# 步骤1: 上传音频文件
|
||||
logger.info("正在上传音频文件...")
|
||||
upload_url = f'https://api.minimax.chat/v1/files/upload?GroupId={group_id}'
|
||||
upload_headers = {'authority': 'api.minimax.chat', 'Authorization': f'Bearer {api_key}'}
|
||||
|
||||
upload_data = {'purpose': 'voice_clone'}
|
||||
|
||||
with open(audio_file_path, 'rb') as audio_file:
|
||||
files = {'file': audio_file}
|
||||
upload_response = requests.post(upload_url, headers=upload_headers, data=upload_data, files=files)
|
||||
|
||||
if upload_response.status_code != 200:
|
||||
logger.error(f"文件上传失败: {upload_response.status_code}")
|
||||
return None, None
|
||||
|
||||
file_id = upload_response.json().get("file", {}).get("file_id")
|
||||
if not file_id:
|
||||
logger.error("无法获取文件ID")
|
||||
return None, None
|
||||
|
||||
logger.success(f"文件上传成功，文件ID: {file_id}")
|
||||
|
||||
# 步骤2: 音频复刻
|
||||
logger.info("正在进行音频复刻...")
|
||||
clone_url = f'https://api.minimax.chat/v1/voice_clone?GroupId={group_id}'
|
||||
clone_payload = json.dumps(
|
||||
{"file_id": file_id, "voice_id": voice_id, "text": text, "model": model, "accuracy": accuracy, })
|
||||
|
||||
clone_headers = {'Authorization': f'Bearer {api_key}', 'content-type': 'application/json'}
|
||||
|
||||
clone_response = requests.post(clone_url, headers=clone_headers, data=clone_payload)
|
||||
|
||||
if clone_response.status_code != 200:
|
||||
logger.error(f"音频复刻失败: {clone_response.status_code}")
|
||||
return None, None
|
||||
|
||||
response_data = clone_response.json()
|
||||
logger.info(f"复刻响应: {response_data}")
|
||||
|
||||
# 获取音频下载链接
|
||||
demo_audio_url = response_data.get("demo_audio")
|
||||
if not demo_audio_url:
|
||||
logger.error("无法获取音频下载链接")
|
||||
return None, None
|
||||
|
||||
# 步骤3: 下载音频文件
|
||||
logger.info("正在下载生成的音频文件...")
|
||||
|
||||
# 创建输出目录
|
||||
if not os.path.exists(output_dir):
|
||||
os.makedirs(output_dir)
|
||||
|
||||
# 使用voice_id作为文件名
|
||||
filename = f"{voice_id}.mp3"
|
||||
output_path = os.path.join(output_dir, filename)
|
||||
|
||||
# 下载文件
|
||||
download_response = requests.get(demo_audio_url)
|
||||
if download_response.status_code != 200:
|
||||
logger.error(f"音频下载失败: {download_response.status_code}")
|
||||
return None, None
|
||||
|
||||
with open(output_path, 'wb') as f:
|
||||
f.write(download_response.content)
|
||||
|
||||
logger.success(f"音频文件下载成功: {output_path}")
|
||||
logger.success(f"使用的voice_id: {voice_id}")
|
||||
return os.path.abspath(output_path), voice_id
|
||||
|
||||
except Exception as e:
|
||||
logger.error(f"处理过程中出现错误: {str(e)}")
|
||||
return None, None
|
||||
|
||||
|
||||
# 使用示例
|
||||
if __name__ == "__main__":
|
||||
# 配置你的参数
|
||||
group_id = "1925197219753894100"
|
||||
api_key = "eyJhbGciOiJSUzI1NiIsInR5cCI6IkpXVCJ9.eyJHcm91cE5hbWUiOiLkuI3lv5jmmbrog73np5HmioDvvIjmna3lt57vvInmnInpmZDlhazlj7giLCJVc2VyTmFtZSI6IuWFiOiSmSIsIkFjY291bnQiOiLlhYjokplAMTkyNTE5NzIxOTc1Mzg5NDEwMCIsIlN1YmplY3RJRCI6IjE5MjUwNjg3Mjg2NTU5NDE2NDQiLCJQaG9uZSI6IiIsIkdyb3VwSUQiOiIxOTI1MTk3MjE5NzUzODk0MTAwIiwiUGFnZU5hbWUiOiIiLCJNYWlsIjoiIiwiQ3JlYXRlVGltZSI6IjIwMjUtMDUtMjIgMTY6MzI6MzkiLCJUb2tlblR5cGUiOjEsImlzcyI6Im1pbmltYXgifQ.ZoyN7nWPLbB-p8Agtw0ggeDd-Bj7T_0KVmFq1g2IKg6_sbmCySqavbZKYBh0AJMIyzxDhYbWXwl8IN_biOPqYjf9QkZFaKftBqn35pHlqWe94zGOD_0T2HYMNH-_WLjpG9NH54GICEN_-oOnYdyjnf74p8AIWR0POvLJ4ijZyz4q22QP51DEb3ExhlwXv68vY-sru33CB9C9e88SR6h1fB1S-eysck_ZenIbfhhXYCqGlAAheQqwft-IS7n1axCqg2Xd3DJj-82Mj1gvlvf2q9CiE8bOu6YvW6Hs67p_60e9Vr8RVEyZc6WzRViqLR1izb8elChmRTdtbsHeq9lSzg"
|
||||
input_audio_path = r'D:\gxm\desktop\minmax_audio_clone_dawan.mp3'
|
||||
text = "我有这么这么多新款我都卖不过来了，我今年我上新到现在，我春夏款，我光裤子类目，300个品光裤子单品有300条，你想一下开发能力这么强的，抖音上面有几家。"
|
||||
output_dir = r'D:\gxm\desktop'
|
||||
|
||||
# 调用函数
|
||||
result_path, used_voice_id = voice_clone_and_download(audio_file_path=input_audio_path, group_id=group_id,
|
||||
api_key=api_key, text=text, output_dir=output_dir)
|
||||
|
||||
Loading…
Reference in New Issue