Update app.py

app.py

@@ -2,6 +2,9 @@ import gradio as gr
 import os
 import shutil
 import tarfile
+import torch
+import torchaudio
+import torchaudio.functional as F
 from huggingface_hub import snapshot_download
 from omegaconf import OmegaConf
 from nemo.collections.asr.models import ASRModel
@@ -68,7 +71,24 @@ def transcribe(audio_path):
         return "Please upload or record audio."
     
     try:
-        transcription = model.transcribe(paths2audio_files=[audio_path])[0]
+        # 1. Load the audio file that Gradio provides
+        waveform, sample_rate = torchaudio.load(audio_path)
+        
+        # 2. Convert to Mono (average the channels if it's stereo)
+        if waveform.shape[0] > 1:
+            waveform = torch.mean(waveform, dim=0, keepdim=True)
+            
+        # 3. Convert to 16000 Hz (Standard for NeMo models)
+        if sample_rate != 16000:
+            waveform = F.resample(waveform, sample_rate, 16000)
+            sample_rate = 16000
+            
+        # 4. Save the cleaned audio to a temporary file
+        processed_path = audio_path + "_mono_16k.wav"
+        torchaudio.save(processed_path, waveform, sample_rate)
+        
+        # 5. Pass the strictly formatted audio to the model
+        transcription = model.transcribe(paths2audio_files=[processed_path])[0]
         
         if isinstance(transcription, list):
             return transcription[0]