app.py

import gradio as gr
from transformers import pipeline, WhisperProcessor, WhisperForConditionalGeneration
import torch
import edge_tts
import asyncio
import numpy as np

class FrenchLearningApp:
    def __init__(self):
        # Initialize models
        self.conversation_model = pipeline("text-generation", model="gpt2")
        
        # Initialize Whisper model
        self.whisper_processor = WhisperProcessor.from_pretrained("openai/whisper-large-v3")
        self.whisper_model = WhisperForConditionalGeneration.from_pretrained("openai/whisper-large-v3")

        self.context = "Start a conversation in French"
        self.learning_goals = []

    def set_learning_goals(self, goals):
        self.learning_goals = goals.split('\n')
        return f"Learning goals set: {self.learning_goals}"

    async def generate_french(self):
        french_text = self.conversation_model(self.context, max_length=100)[0]['generated_text']
        
        # Generate audio using edge-tts
        voice = "fr-FR-HenriNeural"  # You can change this to any available French voice
        communicate = edge_tts.Communicate(french_text, voice)
        audio_data = b""
        async for chunk in communicate.stream():
            if chunk["type"] == "audio":
                audio_data += chunk["data"]
        
        # Convert audio to numpy array for Gradio
        audio_np = np.frombuffer(audio_data, dtype=np.int16)
        audio_float = audio_np.astype(np.float32) / 32768.0  # Convert to float32
        
        return (24000, audio_float), french_text  # 24000 is the default sample rate for edge-tts

    def process_user_response(self, audio):
        # Transcribe audio to French text using Whisper
        input_features = self.whisper_processor(audio, sampling_rate=16000, return_tensors="pt").input_features
        
        # Generate French transcription
        self.whisper_model.config.forced_decoder_ids = self.whisper_processor.get_decoder_prompt_ids(language="french", task="transcribe")
        predicted_ids = self.whisper_model.generate(input_features)
        french_text = self.whisper_processor.batch_decode(predicted_ids, skip_special_tokens=True)[0]

        # Translate French to English using Whisper
        self.whisper_model.config.forced_decoder_ids = self.whisper_processor.get_decoder_prompt_ids(language="french", task="translate")
        predicted_ids = self.whisper_model.generate(input_features)
        english_text = self.whisper_processor.batch_decode(predicted_ids, skip_special_tokens=True)[0]

        # Analyze response (simplified)
        analysis = self.analyze_response(english_text)

        # Update context
        self.context += f" {french_text}"

        return french_text, english_text, analysis

    def analyze_response(self, english_text):
        # Simplified analysis - check if any learning goal is mentioned
        analysis = []
        for goal in self.learning_goals:
            if goal.lower() in english_text.lower():
                analysis.append(f"Goal met: {goal}")
            else:
                analysis.append(f"Goal not yet met: {goal}")
        return "\n".join(analysis)

def launch_app():
    app = FrenchLearningApp()

    with gr.Blocks() as interface:
        gr.Markdown("# French Learning Application")
        
        with gr.Tab("Teacher Setup"):
            goals_input = gr.Textbox(label="Enter learning goals (one per line)")
            set_goals_button = gr.Button("Set Learning Goals")
            goals_output = gr.Textbox(label="Goals Status")
            
            set_goals_button.click(app.set_learning_goals, inputs=goals_input, outputs=goals_output)

        with gr.Tab("Conversation"):
            generate_button = gr.Button("Generate French")
            audio_output = gr.Audio(label="AI Speech")
            french_output = gr.Textbox(label="French Text")

            generate_button.click(lambda: asyncio.run(app.generate_french()), inputs=None, outputs=[audio_output, french_output])

            audio_input = gr.Audio(source="microphone", type="numpy", label="Your Response")
            transcription_output = gr.Textbox(label="Your Speech (Transcribed)")
            translation_output = gr.Textbox(label="English Translation")
            analysis_output = gr.Textbox(label="Analysis")

            audio_input.change(app.process_user_response, inputs=audio_input, 
                               outputs=[transcription_output, translation_output, analysis_output])

    interface.launch()

if __name__ == "__main__":
    launch_app()