v2

app.py

@@ -3,7 +3,7 @@ import whisperx
 import whisper
 
 def transcribe(audio_file):
-    device = "cuda"
+    device = "cuda" if torch.cuda.is_available() else "cpu"
 
     # Transcribe with original Whisper
     model = whisper.load_model("large", device)
@@ -15,14 +15,9 @@ def transcribe(audio_file):
     # Align Whisper output
     result_aligned = whisperx.align(result["segments"], model_a, metadata, audio_file, device)
 
-    return result_aligned["segments"], result_aligned["word_segments"]
+    return {"aligned": result_aligned["segments"], "word_segments": result_aligned["word_segments"]}
 
-# Define Gradio interface
-inputs = gr.inputs.Audio(source="upload", type="file")
-outputs = [
-    gr.outputs.Textbox(label="Segments (before alignment)"),
-    gr.outputs.Textbox(label="Segments (after alignment)"),
-]
+inputs = gr.inputs.Audio(source="upload", type="filepath")
+outputs = gr.outputs.JSON(type="auto")
 
-iface = gr.Interface(fn=transcribe, inputs=inputs, outputs=outputs, title="WhisperX Transcription")
-iface.launch()
+gr.Interface(fn=transcribe, inputs=inputs, outputs=outputs).launch()