Update app.py
Browse files
app.py
CHANGED
|
@@ -8,11 +8,10 @@ from fastapi import FastAPI, UploadFile, File, Form, HTTPException
|
|
| 8 |
from fastapi.responses import JSONResponse, PlainTextResponse
|
| 9 |
from transformers import pipeline
|
| 10 |
|
| 11 |
-
# 1.
|
| 12 |
MODEL_NAME = "openai/whisper-small"
|
| 13 |
-
|
| 14 |
-
# 2. 初始化 Pipeline
|
| 15 |
device = "cuda" if torch.cuda.is_available() else "cpu"
|
|
|
|
| 16 |
pipe = pipeline(
|
| 17 |
"automatic-speech-recognition",
|
| 18 |
model=MODEL_NAME,
|
|
@@ -20,7 +19,7 @@ pipe = pipeline(
|
|
| 20 |
device=device
|
| 21 |
)
|
| 22 |
|
| 23 |
-
#
|
| 24 |
@spaces.GPU
|
| 25 |
def run_whisper_inference(audio_path: str, target_language: str = None, is_translate: bool = False):
|
| 26 |
generate_kwargs = {}
|
|
@@ -34,8 +33,7 @@ def run_whisper_inference(audio_path: str, target_language: str = None, is_trans
|
|
| 34 |
result = pipe(audio_path, generate_kwargs=generate_kwargs)
|
| 35 |
return result["text"]
|
| 36 |
|
| 37 |
-
|
| 38 |
-
# --- Gradio 界面 ---
|
| 39 |
def gradio_predict(audio_path):
|
| 40 |
if audio_path is None:
|
| 41 |
return "请先上传音频或录音!"
|
|
@@ -52,8 +50,7 @@ demo = gr.Interface(
|
|
| 52 |
description="【完美兼容 OpenAI 规范】"
|
| 53 |
)
|
| 54 |
|
| 55 |
-
|
| 56 |
-
# --- FastAPI 接口配置 ---
|
| 57 |
app = FastAPI()
|
| 58 |
|
| 59 |
async def process_openai_audio_request(file: UploadFile, response_format: str, language: str, is_translate: bool):
|
|
@@ -63,7 +60,6 @@ async def process_openai_audio_request(file: UploadFile, response_format: str, l
|
|
| 63 |
temp_path = temp_file.name
|
| 64 |
|
| 65 |
try:
|
| 66 |
-
# 调用最外层带 @spaces.GPU 装饰的函数
|
| 67 |
text = run_whisper_inference(temp_path, target_language=language, is_translate=is_translate)
|
| 68 |
except Exception as e:
|
| 69 |
raise HTTPException(status_code=500, detail=f"Inference failed: {str(e)}")
|
|
@@ -76,7 +72,6 @@ async def process_openai_audio_request(file: UploadFile, response_format: str, l
|
|
| 76 |
|
| 77 |
return JSONResponse(content={"text": text})
|
| 78 |
|
| 79 |
-
|
| 80 |
@app.post("/v1/audio/transcriptions")
|
| 81 |
async def transcribe_api(
|
| 82 |
file: UploadFile = File(...),
|
|
@@ -93,7 +88,6 @@ async def transcribe_api(
|
|
| 93 |
is_translate=False
|
| 94 |
)
|
| 95 |
|
| 96 |
-
|
| 97 |
@app.post("/v1/audio/translations")
|
| 98 |
async def translate_api(
|
| 99 |
file: UploadFile = File(...),
|
|
@@ -109,5 +103,5 @@ async def translate_api(
|
|
| 109 |
is_translate=True
|
| 110 |
)
|
| 111 |
|
| 112 |
-
# 挂载 Gradio
|
| 113 |
app = gr.mount_gradio_app(app, demo, path="/")
|
|
|
|
| 8 |
from fastapi.responses import JSONResponse, PlainTextResponse
|
| 9 |
from transformers import pipeline
|
| 10 |
|
| 11 |
+
# 1. 模型加载
|
| 12 |
MODEL_NAME = "openai/whisper-small"
|
|
|
|
|
|
|
| 13 |
device = "cuda" if torch.cuda.is_available() else "cpu"
|
| 14 |
+
|
| 15 |
pipe = pipeline(
|
| 16 |
"automatic-speech-recognition",
|
| 17 |
model=MODEL_NAME,
|
|
|
|
| 19 |
device=device
|
| 20 |
)
|
| 21 |
|
| 22 |
+
# 2. 核心 GPU 推理函数(ZeroGPU 严格要求必须在文件顶层声明)
|
| 23 |
@spaces.GPU
|
| 24 |
def run_whisper_inference(audio_path: str, target_language: str = None, is_translate: bool = False):
|
| 25 |
generate_kwargs = {}
|
|
|
|
| 33 |
result = pipe(audio_path, generate_kwargs=generate_kwargs)
|
| 34 |
return result["text"]
|
| 35 |
|
| 36 |
+
# 3. Gradio UI 封装
|
|
|
|
| 37 |
def gradio_predict(audio_path):
|
| 38 |
if audio_path is None:
|
| 39 |
return "请先上传音频或录音!"
|
|
|
|
| 50 |
description="【完美兼容 OpenAI 规范】"
|
| 51 |
)
|
| 52 |
|
| 53 |
+
# 4. FastAPI 应用声明
|
|
|
|
| 54 |
app = FastAPI()
|
| 55 |
|
| 56 |
async def process_openai_audio_request(file: UploadFile, response_format: str, language: str, is_translate: bool):
|
|
|
|
| 60 |
temp_path = temp_file.name
|
| 61 |
|
| 62 |
try:
|
|
|
|
| 63 |
text = run_whisper_inference(temp_path, target_language=language, is_translate=is_translate)
|
| 64 |
except Exception as e:
|
| 65 |
raise HTTPException(status_code=500, detail=f"Inference failed: {str(e)}")
|
|
|
|
| 72 |
|
| 73 |
return JSONResponse(content={"text": text})
|
| 74 |
|
|
|
|
| 75 |
@app.post("/v1/audio/transcriptions")
|
| 76 |
async def transcribe_api(
|
| 77 |
file: UploadFile = File(...),
|
|
|
|
| 88 |
is_translate=False
|
| 89 |
)
|
| 90 |
|
|
|
|
| 91 |
@app.post("/v1/audio/translations")
|
| 92 |
async def translate_api(
|
| 93 |
file: UploadFile = File(...),
|
|
|
|
| 103 |
is_translate=True
|
| 104 |
)
|
| 105 |
|
| 106 |
+
# 5. 挂载 Gradio 到 FastAPI
|
| 107 |
app = gr.mount_gradio_app(app, demo, path="/")
|