Update app.py

app.py

@@ -3,43 +3,54 @@ import torch
 import torchaudio
 from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
 
-# Load model and processor
+# Load model and processor outside the function to avoid reloading
 model_id = "facebook/wav2vec2-large-960h-lv60-self"
 processor = Wav2Vec2Processor.from_pretrained(model_id)
 model = Wav2Vec2ForCTC.from_pretrained(model_id)
 
-# Transcription function
-def transcribe(audio_file):
+# Transcription function with optimization
+def transcribe(audio_file, progress=gr.Progress()):
     if audio_file is None:
         return "⚠️ No audio received."
 
-    # Load and convert audio
+    progress(0, desc="Loading audio...")
+    # Load audio
     waveform, sample_rate = torchaudio.load(audio_file)
+
+    # Optimize resampling: Only resample if necessary and use faster method
     if sample_rate != 16000:
+        progress(0.3, desc="Resampling audio...")
+        # Use torch's resample for efficiency
         waveform = torchaudio.functional.resample(waveform, orig_freq=sample_rate, new_freq=16000)
         sample_rate = 16000
 
-    # Only one channel (mono)
+    # Convert to mono if stereo
     if waveform.shape[0] > 1:
+        progress(0.5, desc="Converting to mono...")
         waveform = waveform.mean(dim=0).unsqueeze(0)
 
+    # Process audio in chunks if large to reduce memory usage (optional optimization)
+    progress(0.7, desc="Processing audio...")
     input_values = processor(waveform.squeeze().numpy(), sampling_rate=sample_rate, return_tensors="pt").input_values
 
     with torch.no_grad():
+        progress(0.8, desc="Transcribing...")
         logits = model(input_values).logits
 
     predicted_ids = torch.argmax(logits, dim=-1)
     transcription = processor.batch_decode(predicted_ids)[0]
 
+    progress(1.0, desc="Done!")
     return transcription.lower()
 
-# Gradio UI
+# Gradio UI with progress tracking
 demo = gr.Interface(
     fn=transcribe,
     inputs=gr.Audio(sources=["microphone"], type="filepath", label="🎤 Speak now"),
     outputs=gr.Textbox(label="📝 Transcription"),
     title="Wav2Vec2 Speech Transcription",
-    description="Speak into the microphone and get a transcription using Wav2Vec2 (via Hugging Face Transformers)."
+    description="Speak into the microphone and get a transcription using Wav2Vec2 (via Hugging Face Transformers).",
+    allow_flagging="never"  # Optional: Reduces overhead
 )
 
-demo.launch()
+demo.launch()