Update app.py

app.py

@@ -8,6 +8,11 @@ import os, re, json
 import gradio as gr
 import google.generativeai as genai
 
+import onnxruntime as ort
+import numpy as np
+import soundfile as sf
+
+
 # Safe API configuration
 api_key = os.environ.get("GEMINI_API_KEY")
 if api_key:
@@ -37,6 +42,66 @@ SAMPLES = {
         "for 15 days",
 }
 
+
+MODEL_PATH = "model_quantized.onnx"  # adjust if different
+
+session = None
+
+def load_model():
+    global session
+    if session is None:
+        session = ort.InferenceSession(MODEL_PATH)
+    return session
+
+def preprocess_audio(audio_path):
+    audio, sr = sf.read(audio_path)
+
+    if len(audio.shape) > 1:
+        audio = np.mean(audio, axis=1)
+
+    # Resample if needed
+    if sr != 16000:
+        from scipy.signal import resample
+        audio = resample(audio, int(len(audio) * 16000 / sr))
+
+    return audio.astype(np.float32)
+
+
+def transcribe_audio(audio_path):
+    try:
+        session = load_model()
+        audio = preprocess_audio(audio_path)
+
+        inputs = {session.get_inputs()[0].name: audio}
+
+        outputs = session.run(None, inputs)
+
+        # This part depends on your model decoding
+        transcript = str(outputs[0])
+
+        return transcript
+
+    except Exception as e:
+        return f"ASR Error: {e}"
+
+
+def run_full_pipeline(audio):
+    if not audio:
+        return "No audio", "", "", ""
+
+    transcript = transcribe_audio(audio)
+
+    if transcript.startswith("ASR Error"):
+        return transcript, "", "", ""
+
+    correction = run_correction(transcript)
+    soap = run_soap(transcript)
+    entities = run_entities(transcript)
+
+    return transcript, correction, soap, entities
+
+
+
 def prompt_correction(t):
     return f"""You are a medical transcript editor.
 
@@ -146,6 +211,12 @@ with gr.Blocks(
     theme=gr.themes.Soft(),
 ) as demo:
 
+
+    audio_input = gr.Audio(type="filepath", label="Upload Medical Audio")
+
+    run_audio_btn = gr.Button("🎤 Run Full Pipeline")
+    raw_transcript_out = gr.Textbox(label="Raw Transcript", lines=6)
+    
     gr.Markdown("""
     # ⚕ MedASR + Gemini Pipeline
     Phase 2 — Transcript → Clinical NLP (Correction, SOAP, Entities)
@@ -195,6 +266,11 @@ with gr.Blocks(
         inputs=transcript_box,
         outputs=[correction_out, soap_out, entities_out]
     )
+    run_audio_btn.click(
+    fn=run_full_pipeline,
+    inputs=audio_input,
+    outputs=[raw_transcript_out, correction_out, soap_out, entities_out]
+)
 
 if __name__ == "__main__":
     demo.launch()