Update app.py

app.py

@@ -5,75 +5,79 @@ from TTS.api import TTS
 import tempfile
 import os
 import speech_recognition as sr
+from difflib import SequenceMatcher
 
-# Load question generation model
+# Load models
 qg_pipeline = pipeline("text2text-generation", model="valhalla/t5-small-e2e-qg")
-
-# Load TTS model
 tts = TTS(model_name="tts_models/en/ljspeech/tacotron2-DDC", progress_bar=False, gpu=False)
 
-# Global storage
-last_answer = ""
-
-def generate_question(text):
-    global last_answer
-
-    # Extract a possible answer from the text (you can improve this logic)
-    last_answer = "They are Aladdin lamps"  # You can auto-extract later
+# Simulate QA by extracting key sentence from input text (placeholder for real QA)
+def extract_answer(question, context):
+    for line in context.split("\n"):
+        if any(word.lower() in line.lower() for word in question.split()[:3]):
+            return line
+    return ""
 
-    # Prompt for question generation
-    input_text = f"generate question: {text.strip()}"
-    generated = qg_pipeline(input_text, max_length=64)[0]["generated_text"]
+def generate_questions(text):
+    output = qg_pipeline(f"generate questions: {text}", num_return_sequences=3)
+    questions = [q["generated_text"] for q in output]
+    return questions, text, 0  # store context and index
 
-    # Keep only the first question
-    question = generated.split("<sep>")[0].strip()
+def ask_question(state):
+    questions, context, idx = state
+    if idx >= len(questions):
+        return "All questions asked.", None, state
 
-    # Generate audio
+    question = questions[idx]
     with tempfile.NamedTemporaryFile(delete=False, suffix=".wav") as fp:
         tts.tts_to_file(text=question, file_path=fp.name)
         audio_path = fp.name
 
-    return question, audio_path
+    return question, audio_path, (questions, context, idx)
+
+def transcribe_and_feedback(audio_path, state):
+    questions, context, idx = state
+    if idx == 0 or idx > len(questions):
+        return "Please ask a question first.", state
 
-def transcribe_and_check(audio_path):
-    global expected_answer
     recognizer = sr.Recognizer()
     with sr.AudioFile(audio_path) as source:
         audio_data = recognizer.record(source)
         try:
-            user_answer = recognizer.recognize_google(audio_data).lower()
-        except sr.UnknownValueError:
-            return "Sorry, I couldn't understand your answer."
-        except sr.RequestError:
-            return "Speech recognition error."
-
-    # Simple keyword-based matching
-    user_words = set(user_answer.split())
-    expected_words = set(expected_answer.lower().split())
-
-    common = user_words.intersection(expected_words)
-    score = len(common) / max(1, len(expected_words))
-
-    if score > 0.3 or any(word in user_answer for word in ["special", "lamp"]):
-        return f"✅ Good answer: '{user_answer}'"
+            user_answer = recognizer.recognize_google(audio_data)
+        except:
+            return "Could not understand the answer.", state
+
+    # Simulate expected answer
+    question = questions[idx - 1]
+    expected = extract_answer(question, context)
+    ratio = SequenceMatcher(None, user_answer.lower(), expected.lower()).ratio()
+    if ratio > 0.6:
+        feedback = f"✅ Good answer: {user_answer}"
     else:
-        return f"❌ Try again. You said: '{user_answer}'"
+        feedback = f"❌ Try again. You said: {user_answer}"
+
+    return feedback, (questions, context, idx)
 
 with gr.Blocks() as app:
-    gr.Markdown("### 🎓 Interactive Coursebook Q&A")
+    gr.Markdown("### 🎓 Interactive Q&A Lesson")
+
+    with gr.Row():
+        course_text = gr.Textbox(lines=8, label="Paste Coursebook Text")
+        gen_btn = gr.Button("🔄 Generate Questions")
 
-    course_text = gr.Textbox(lines=6, label="📘 Coursebook Text")
+    question_text = gr.Textbox(label="Current Question")
+    question_audio = gr.Audio(label="Listen to Question", type="filepath")
+    ask_btn = gr.Button("▶️ Ask Next Question")
 
-    generate_btn = gr.Button("🔊 Generate Question and Speak")
-    question_output = gr.Textbox(label="🧠 Generated Question")
-    audio_output = gr.Audio(label="🔈 Question Audio", type="filepath")
+    user_audio = gr.Audio(label="Your Spoken Answer", sources="microphone", type="filepath")
+    transcribe_btn = gr.Button("📝 Submit Answer")
+    feedback_output = gr.Textbox(label="Feedback")
 
-    user_audio = gr.Audio(label="🎤 Your Answer", type="filepath", sources=["microphone"])
-    transcribe_btn = gr.Button("📝 Transcribe Answer")
-    transcription_output = gr.Textbox(label="🗣 Transcribed Answer")
-    feedback_output = gr.Textbox(label="🧪 Feedback")
+    conversation_state = gr.State()
 
-    generate_btn.click(fn=generate_question, inputs=course_text, outputs=[question_output, audio_output])
-    transcribe_btn.click(fn=transcribe_and_check, inputs=user_audio, outputs=[transcription_output, feedback_output])
+    gen_btn.click(generate_questions, inputs=course_text, outputs=[conversation_state, course_text, gr.State(0)])
+    ask_btn.click(ask_question, inputs=conversation_state, outputs=[question_text, question_audio, conversation_state])
+    transcribe_btn.click(transcribe_and_feedback, inputs=[user_audio, conversation_state], outputs=[feedback_output, conversation_state])
 
 app.launch()