Update app.py

app.py

@@ -1,23 +1,18 @@
 import gradio as gr
-from transformers import pipeline, AutoModelForSeq2SeqLM, AutoTokenizer, WhisperProcessor, WhisperForConditionalGeneration
+from transformers import pipeline, WhisperProcessor, WhisperForConditionalGeneration
 import torch
 import edge_tts
 import asyncio
-import soundfile as sf
-import io
 import numpy as np
 
 class FrenchLearningApp:
     def __init__(self):
         # Initialize models
         self.conversation_model = pipeline("text-generation", model="gpt2")
-        self.translation_model = AutoModelForSeq2SeqLM.from_pretrained("facebook/mbart-large-50-many-to-many-mmt")
-        self.translation_tokenizer = AutoTokenizer.from_pretrained("facebook/mbart-large-50-many-to-many-mmt")
         
         # Initialize Whisper model
         self.whisper_processor = WhisperProcessor.from_pretrained("openai/whisper-large-v3")
         self.whisper_model = WhisperForConditionalGeneration.from_pretrained("openai/whisper-large-v3")
-        self.whisper_model.config.forced_decoder_ids = self.whisper_processor.get_decoder_prompt_ids(language="french", task="transcribe")
 
         self.context = "Start a conversation in French"
         self.learning_goals = []
@@ -44,15 +39,18 @@ class FrenchLearningApp:
         return (24000, audio_float), french_text  # 24000 is the default sample rate for edge-tts
 
     def process_user_response(self, audio):
-        # Transcribe audio to text using Whisper
+        # 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
-        inputs = self.translation_tokenizer(french_text, return_tensors="pt")
-        translated = self.translation_model.generate(**inputs)
-        english_text = self.translation_tokenizer.decode(translated[0], skip_special_tokens=True)
+        # 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)
@@ -103,4 +101,4 @@ def launch_app():
     interface.launch()
 
 if __name__ == "__main__":
-    launch_app()
+    launch_app()