Update app.py

import gradio as gr
from transformers import pipeline
import torch
from TTS.api import TTS
import tempfile
import os
import speech_recognition as sr
from difflib import SequenceMatcher

# Load models
qg_pipeline = pipeline("text2text-generation", model="valhalla/t5-small-e2e-qg")
tts = TTS(model_name="tts_models/en/ljspeech/tacotron2-DDC", progress_bar=False, gpu=False)

# Simulate QA by extracting key sentence from input text (placeholder)
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 ""

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) # this tuple is stored in state

def ask_question(state):
questions, context, idx = state
if idx >= len(questions):
return "✅ All questions asked.", None, state

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, (questions, context, idx + 1)

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

recognizer = sr.Recognizer()
with sr.AudioFile(audio_path) as source:
audio_data = recognizer.record(source)
try:
user_answer = recognizer.recognize_google(audio_data)
except:
return "❌ Could not understand your answer.", state

# Compare with expected answer
question = questions[idx - 1] # subtract 1 because idx was already incremented
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:
feedback = f"❌ Try again. You said: {user_answer}"

return feedback, (questions, context, idx)

with gr.Blocks() as app:
gr.Markdown("### 🎓 Interactive Speaking Practice with Coursebook Dialogues")

with gr.Row():
course_text = gr.Textbox(lines=8, label="📘 Paste Coursebook Text")
gen_btn = gr.Button("🔄 Generate Questions")

question_text = gr.Textbox(label="🎙️ Current Question")
question_audio = gr.Audio(label="🔊 Listen to Question", type="filepath")
ask_btn = gr.Button("▶️ Ask Next Question")

user_audio = gr.Audio(label="🎧 Your Spoken Answer", sources="microphone", type="filepath")
transcribe_btn = gr.Button("📝 Submit Answer")
feedback_output = gr.Textbox(label="🗨️ Feedback")

conversation_state = gr.State()

gen_btn.click(fn=generate_questions, inputs=course_text, outputs=conversation_state)
ask_btn.click(fn=ask_question, inputs=conversation_state, outputs=[question_text, question_audio, conversation_state])
transcribe_btn.click(fn=transcribe_and_feedback, inputs=[user_audio, conversation_state], outputs=[feedback_output, conversation_state])

app.launch()

app.py

@@ -1,83 +0,0 @@
-import gradio as gr
-from transformers import pipeline
-import torch
-from TTS.api import TTS
-import tempfile
-import os
-import speech_recognition as sr
-from difflib import SequenceMatcher
-
-# Load models
-qg_pipeline = pipeline("text2text-generation", model="valhalla/t5-small-e2e-qg")
-tts = TTS(model_name="tts_models/en/ljspeech/tacotron2-DDC", progress_bar=False, gpu=False)
-
-# 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 ""
-
-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
-
-def ask_question(state):
-    questions, context, idx = state
-    if idx >= len(questions):
-        return "All questions asked.", None, state
-
-    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, (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
-
-    recognizer = sr.Recognizer()
-    with sr.AudioFile(audio_path) as source:
-        audio_data = recognizer.record(source)
-        try:
-            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:
-        feedback = f"❌ Try again. You said: {user_answer}"
-
-    return feedback, (questions, context, idx)
-
-with gr.Blocks() as app:
-    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")
-
-    question_text = gr.Textbox(label="Current Question")
-    question_audio = gr.Audio(label="Listen to Question", type="filepath")
-    ask_btn = gr.Button("▶️ Ask Next Question")
-
-    user_audio = gr.Audio(label="Your Spoken Answer", sources="microphone", type="filepath")
-    transcribe_btn = gr.Button("📝 Submit Answer")
-    feedback_output = gr.Textbox(label="Feedback")
-
-    conversation_state = gr.State()
-
-    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()