Fix torch import error in translate-audio endpoint

app.py

@@ -6,8 +6,10 @@ import logging
 import threading
 import tempfile
 import uuid
+import torch
 import numpy as np
 import soundfile as sf
+import librosa
 from fastapi import FastAPI, HTTPException, UploadFile, File, Form
 from fastapi.responses import JSONResponse
 from typing import Dict, Any, Optional
@@ -50,12 +52,10 @@ def load_models_task():
     try:
         loading_in_progress = True
         
-        # Import heavy libraries only when needed
+        # Load STT model
         logger.info("Starting to load STT model...")
-        import torch
         from transformers import WhisperProcessor, WhisperForConditionalGeneration
         
-        # Load STT model
         try:
             logger.info("Loading Whisper model...")
             model_status["stt"] = "loading"
@@ -177,15 +177,18 @@ async def translate_audio(audio: UploadFile = File(...), source_lang: str = Form
     
     try:
         # Read and preprocess the audio
+        logger.info(f"Reading audio file: {temp_path}")
         waveform, sample_rate = sf.read(temp_path)
+        logger.info(f"Audio loaded: sample_rate={sample_rate}, waveform_shape={waveform.shape}")
         if sample_rate != 16000:
             logger.info(f"Resampling audio from {sample_rate} Hz to 16000 Hz")
-            import librosa
             waveform = librosa.resample(waveform, orig_sr=sample_rate, target_sr=16000)
         
         # Process the audio with Whisper
         device = "cuda" if torch.cuda.is_available() else "cpu"
+        logger.info(f"Using device: {device}")
         inputs = stt_processor(waveform, sampling_rate=16000, return_tensors="pt").to(device)
+        logger.info("Audio processed, generating transcription...")
         with torch.no_grad():
             generated_ids = stt_model.generate(**inputs)
             transcription = stt_processor.batch_decode(generated_ids, skip_special_tokens=True)[0]
@@ -210,6 +213,7 @@ async def translate_audio(audio: UploadFile = File(...), source_lang: str = Form
             "output_audio": None
         }
     finally:
+        logger.info(f"Cleaning up temporary file: {temp_path}")
         os.unlink(temp_path)
 
 if __name__ == "__main__":