Update app.py

app.py

@@ -1,18 +1,20 @@
 import gradio as gr
 import os
 import uuid
-from gtts import gTTS
 import whisper
+from gtts import gTTS
+import subprocess
 import openai
-import wave
 
-# Load models once
-# Force whisper to use a writable cache dir
+# Use writable cache for Hugging Face
 os.environ["XDG_CACHE_HOME"] = "/tmp/.cache"
+os.environ["MPLCONFIGDIR"] = "/tmp/matplotlib"
+
+# Load Whisper and set OpenAI key
 model = whisper.load_model("base")
 openai.api_key = os.getenv("OPENAI_API_KEY")
 
-# LLM prompt template
+# Prompt template
 PROMPT_TEMPLATE = """
 You're a communication coach helping users improve their spoken English.
 
@@ -33,9 +35,16 @@ Transcript:
 Provide concise feedback in 3-4 bullet points followed by a one-line motivational sentence.
 """
 
+# Convert uploaded audio to 16kHz mono WAV
+def convert_to_wav(input_file):
+    output_wav = f"/tmp/{uuid.uuid4()}.wav"
+    command = ["ffmpeg", "-y", "-i", input_file, "-ar", "16000", "-ac", "1", output_wav]
+    subprocess.run(command, stdout=subprocess.DEVNULL, stderr=subprocess.DEVNULL)
+    return output_wav
+
+# Call GPT to analyze transcript
 def generate_feedback_with_llm(transcript):
     prompt = PROMPT_TEMPLATE.format(transcript=transcript.strip())
-
     response = openai.ChatCompletion.create(
         model="gpt-3.5-turbo",
         temperature=0.7,
@@ -44,35 +53,34 @@ def generate_feedback_with_llm(transcript):
             {"role": "user", "content": prompt}
         ]
     )
-
     return response.choices[0].message["content"]
 
-def is_valid_audio(filepath):
-    try:
-        with wave.open(filepath, 'rb') as wf:
-            return wf.getnframes() > 0
-    except Exception:
-        return False
-
+# Main function
 def tutor_feedback(audio_file):
-    if not audio_file or not os.path.exists(audio_file) or not is_valid_audio(audio_file):
-        return "No valid audio received.", "Please record again.", None
-        
-    # Step 1: Transcribe
-    result = model.transcribe(audio_file)
+    if not audio_file or not os.path.exists(audio_file):
+        return "No audio received.", "Please upload or record again.", None
+
+    print(f">>> Audio received: {audio_file}")
+    print(f">>> File size: {os.path.getsize(audio_file)} bytes")
+
+    # Convert to whisper-friendly WAV
+    wav_path = convert_to_wav(audio_file)
+
+    # Transcribe
+    result = model.transcribe(wav_path)
     transcript = result["text"]
 
-    # Step 2: Analyze via LLM
+    # Feedback via GPT
     feedback_text = generate_feedback_with_llm(transcript)
 
-    # Step 3: TTS
+    # TTS
     tts = gTTS(feedback_text)
-    output_path = f"/tmp/{uuid.uuid4()}.mp3"
-    tts.save(output_path)
+    mp3_path = f"/tmp/{uuid.uuid4()}.mp3"
+    tts.save(mp3_path)
 
-    return transcript, feedback_text, output_path
+    return transcript, feedback_text, mp3_path
 
-# Gradio UI
+# Gradio interface
 iface = gr.Interface(
     fn=tutor_feedback,
     inputs=gr.Audio(type="filepath", label="🎤 Upload or record your response"),
@@ -81,10 +89,11 @@ iface = gr.Interface(
         gr.Textbox(label="📢 Tutor Feedback"),
         gr.Audio(label="🔊 Spoken Feedback", type="filepath")
     ],
-    title="🗣 Communications Tutor with LLM Feedback",
-    description="Speak your answer. The AI tutor will transcribe it, analyze it using GPT, and give spoken feedback on how to improve.",
-    allow_flagging="never"  # 👈 This disables flagging entirely
+    title="🗣 Communications Tutor with Whisper + GPT",
+    description="Speak or upload a response. The tutor will transcribe it, analyze with GPT, and respond with spoken feedback.",
+    allow_flagging="never"
 )
 
-print(">>> App is ready. Launching Gradio...")
-iface.launch(server_name="0.0.0.0", server_port=7860, debug=True)
+if __name__ == "__main__":
+    print("✅ App is launching...")
+    iface.launch(server_name="0.0.0.0", server_port=7860, debug=True)