app.py

import whisper as openai_whisper
from transformers import AutoModelForCausalLM, AutoTokenizer
from TTS.api import TTS
import gradio as gr
import torch
import os

# 1. Speech-to-Text (STT) Implementation
def setup_stt():
    model = openai_whisper.load_model("base")  # Explicit OpenAI Whisper
    return model

def transcribe_audio(model, audio_file):
    result = model.transcribe(audio_file)
    print("Transcription:", result['text'])
    return result['text']

# 2. Natural Language Processing (NLP) Implementation
def setup_nlp():
    model_name = "gpt2"
    tokenizer = AutoTokenizer.from_pretrained(model_name)
    model = AutoModelForCausalLM.from_pretrained(model_name)
    return tokenizer, model

def generate_response(tokenizer, model, input_text):
    prompt = f"User: {input_text}\nAssistant:"
    input_ids = tokenizer.encode(prompt, return_tensors="pt")
    
    response = model.generate(
        input_ids,
        max_length=150,
        num_return_sequences=1,
        temperature=0.7,
        top_p=0.9,
        pad_token_id=tokenizer.eos_token_id,
        no_repeat_ngram_size=2
    )
    return tokenizer.decode(response[0], skip_special_tokens=True)

# 3. Text-to-Speech (TTS) Implementation
def setup_tts():
    tts = TTS(model_name="tts_models/en/ljspeech/tacotron2-DDC")
    return tts

def generate_speech(tts, text, file_path="output.wav"):
    tts.tts_to_file(text, file_path=file_path)
    return file_path

# 4. Voice AI System Class
class VoiceAISystem:
    def __init__(self):
        print("Initializing Voice AI System...")
        print("Loading STT model...")
        self.stt_model = setup_stt()
        print("Loading NLP model...")
        self.tokenizer, self.nlp_model = setup_nlp()
        print("Loading TTS model...")
        self.tts_model = setup_tts()
        
        # GPU Optimization
        self.device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
        print(f"Using device: {self.device}")
        self.nlp_model = self.nlp_model.to(self.device)
        print("System initialization complete!")

    def process_audio(self, audio_file):
        try:
            os.makedirs("tmp", exist_ok=True)
            
            print("Transcribing audio...")
            text = transcribe_audio(self.stt_model, audio_file)
            
            print("Generating response...")
            with torch.cuda.amp.autocast(enabled=torch.cuda.is_available()):
                response = generate_response(self.tokenizer, self.nlp_model, text)
            
            print("Converting response to speech...")
            output_path = os.path.join("tmp", "response.wav")
            audio_response = generate_speech(self.tts_model, response, output_path)
            
            return audio_response, text, response
        except Exception as e:
            print(f"Error during processing: {str(e)}")
            return None, f"Error: {str(e)}", "Error processing request"

# 5. Gradio UI Integration
def create_voice_ai_interface():
    system = VoiceAISystem()
    
    def chat(audio):
        if audio is None:
            return None, "No audio provided", "No response generated"
        return system.process_audio(audio)
    
    interface = gr.Interface(
        fn=chat,
        inputs=[
            gr.Audio(
                type="filepath",
                label="Speak here"
            )
        ],
        outputs=[
            gr.Audio(label="AI Response"),
            gr.Textbox(label="Transcribed Text"),
            gr.Textbox(label="AI Response Text")
        ],
        title="Voice AI System",
        description="Click to record your voice and interact with the AI"
    )
    
    return interface

# Launch the interface
if __name__ == "__main__":
    iface = create_voice_ai_interface()
    iface.launch(share=True)