Upload api.py with huggingface_hub

api.py

@@ -4,6 +4,7 @@ import torch
 import torchaudio
 import numpy as np
 from fastapi import FastAPI, UploadFile, File, Header, HTTPException, status, Form
+from fastapi.responses import PlainTextResponse
 from funasr import AutoModel
 from funasr.utils.postprocess_utils import rich_transcription_postprocess
 
@@ -21,7 +22,7 @@ VAD_MODEL_LOCAL_PATH = os.path.join(MODEL_CACHE_DIR, "fsmn-vad")
 device = "cpu"
 
 print(f"Loading SenseVoice model on {device}...")
-model = AutoModel(
+asr_model = AutoModel(
     model=SENSE_VOICE_SMALL_LOCAL_PATH,
     trust_remote_code=False,
     vad_model=VAD_MODEL_LOCAL_PATH,
@@ -33,6 +34,12 @@ model = AutoModel(
 print("Model loaded successfully.")
 
 
+@app.get("/")
+def read_root():
+    """Hugging Face Space health check endpoint."""
+    return {"status": "ok", "message": "SenseVoice ASR API is running"}
+
+
 @app.post("/transcribe")
 async def transcribe(
     file: UploadFile = File(...),
@@ -59,7 +66,7 @@ async def transcribe(
 
     try:
         # Generate transcription
-        res = model.generate(
+        res = asr_model.generate(
             input=audio_path,
             cache={},
             language="auto",
@@ -93,7 +100,7 @@ async def openai_audio_transcriptions(
 
     Request (multipart/form-data):
       - file: audio file
-      - model: optional model identifier
+      - model: optional model identifier (ignored, uses SenseVoice)
       - response_format: json | text | srt | vtt (default json)
 
     Response:
@@ -118,7 +125,7 @@ async def openai_audio_transcriptions(
         audio_path = tmp.name
 
     try:
-        res = model.generate(
+        res = asr_model.generate(
             input=audio_path,
             cache={},
             language="auto",
@@ -138,7 +145,6 @@ async def openai_audio_transcriptions(
             os.unlink(audio_path)
 
     if fmt == "text":
-        from fastapi.responses import PlainTextResponse
         return PlainTextResponse(content=text)
 
     return {"text": text}