app.py

# import subprocess

# # Install required libraries
# subprocess.check_call(["pip", "install", "torch>=1.11.0"])
# subprocess.check_call(["pip", "install", "transformers>=4.31.0"])
# subprocess.check_call(["pip", "install", "diffusers>=0.14.0"])
# subprocess.check_call(["pip", "install", "librosa"])
# subprocess.check_call(["pip", "install", "accelerate>=0.20.1"])
# subprocess.check_call(["pip", "install", "gradio>=3.35.2"])
# subprocess.check_call(["pip", "install", "huggingface_hub"])

# import os
# import threading
# import numpy as np
# import librosa
# import torch
# import gradio as gr
# from functools import lru_cache
# from transformers import pipeline, AutoModelForCausalLM, AutoTokenizer
# from huggingface_hub import login
# from diffusers import StableDiffusionPipeline, DPMSolverMultistepScheduler

# # Ensure required dependencies are installed
# def install_missing_packages():
#     required_packages = {
#         "librosa": None,
#         "diffusers": ">=0.14.0",
#         "gradio": ">=3.35.2",
#         "huggingface_hub": None,
#         "accelerate": ">=0.20.1",
#         "transformers": ">=4.31.0"
#     }
#     for package, version in required_packages.items():
#         try:
#             __import__(package)
#         except ImportError:
#             package_name = f"{package}{version}" if version else package
#             subprocess.check_call(["pip", "install", package_name])

# install_missing_packages()

# # Get Hugging Face token for authentication
# hf_token = os.getenv("HF_TOKEN")
# if hf_token:
#     login(hf_token)
# else:
#     raise ValueError("HF_TOKEN environment variable not set.")

# # Load speech-to-text model (Whisper)
# speech_to_text = pipeline(
#     "automatic-speech-recognition", 
#     model="openai/whisper-tiny", 
#     return_timestamps=True
# )

# # Load Stable Diffusion model for text-to-image
# text_to_image = StableDiffusionPipeline.from_pretrained("runwayml/stable-diffusion-v1-5")
# device = "cuda" if torch.cuda.is_available() else "cpu"
# text_to_image.to(device)
# text_to_image.enable_attention_slicing()
# text_to_image.safety_checker = None
# text_to_image.scheduler = DPMSolverMultistepScheduler.from_config(text_to_image.scheduler.config)

# # Load ChatGPT-like conversational model
# chat_model_name = "microsoft/DialoGPT-medium"
# chat_tokenizer = AutoTokenizer.from_pretrained(chat_model_name)
# chat_model = AutoModelForCausalLM.from_pretrained(chat_model_name)

# # Preprocess audio file into NumPy array
# def preprocess_audio(audio_path):
#     try:
#         audio, sr = librosa.load(audio_path, sr=16000)  # Resample to 16kHz
#         return np.array(audio, dtype=np.float32)
#     except Exception as e:
#         return f"Error in preprocessing audio: {str(e)}"

# # Speech-to-text function with long-form transcription support
# @lru_cache(maxsize=10)
# def transcribe_audio(audio_path):
#     try:
#         audio_array = preprocess_audio(audio_path)
#         if isinstance(audio_array, str):  # Error message from preprocessing
#             return audio_array
#         result = speech_to_text(audio_array)
#         # Combine text from multiple segments for long-form transcription
#         transcription = " ".join(segment["text"] for segment in result["chunks"])
#         return transcription
#     except Exception as e:
#         return f"Error in transcription: {str(e)}"

# # Text-to-image function
# @lru_cache(maxsize=10)
# def generate_image_from_text(text):
#     try:
#         image = text_to_image(text, height=256, width=256).images[0]  # Generate smaller images for speed
#         return image
#     except Exception as e:
#         return f"Error in image generation: {str(e)}"

# # ChatGPT-like conversational response
# def chat_with_gpt(prompt):
#     try:
#         inputs = chat_tokenizer.encode(prompt, return_tensors="pt")
#         outputs = chat_model.generate(inputs, max_length=200, pad_token_id=chat_tokenizer.eos_token_id)
#         response = chat_tokenizer.decode(outputs[0], skip_special_tokens=True)
#         return response
#     except Exception as e:
#         return f"Error in chat response: {str(e)}"

# # Combined processing function
# def process_audio_and_generate_results(audio_path):
#     transcription_result = {"result": None}
#     image_result = {"result": None}

#     # Function to run transcription and image generation in parallel
#     def transcription_thread():
#         transcription_result["result"] = transcribe_audio(audio_path)

#     def image_generation_thread():
#         transcription = transcription_result["result"]
#         if transcription and "Error" not in transcription:
#             image_result["result"] = generate_image_from_text(transcription)

#     # Start both tasks in parallel
#     t1 = threading.Thread(target=transcription_thread)
#     t2 = threading.Thread(target=image_generation_thread)

#     t1.start()
#     t2.start()

#     t1.join()  # Wait for transcription to finish
#     t2.join()  # Wait for image generation to finish

#     transcription = transcription_result["result"]
#     image = image_result["result"]

#     if "Error" in transcription:
#         return None, transcription
#     if isinstance(image, str) and "Error" in image:
#         return None, image

#     return image, transcription

# # Gradio interface for speech-to-text
# speech_to_text_iface = gr.Interface(
#     fn=transcribe_audio,
#     inputs=gr.Audio(type="filepath", label="Upload audio file for transcription (WAV/MP3)"),
#     outputs=gr.Textbox(label="Transcription"),
#     title="Speech-to-Text Transcription",
#     description="Upload an audio file to transcribe speech into text.",
# )

# # Gradio interface for voice-to-image and chat
# voice_to_image_and_chat_iface = gr.Interface(
#     fn=process_audio_and_generate_results,
#     inputs=gr.Audio(type="filepath", label="Upload audio file (WAV/MP3)"),
#     outputs=[gr.Image(label="Generated Image"), gr.Textbox(label="Transcription")],
#     title="Voice-to-Image and Chat",
#     description="Upload an audio file to transcribe speech to text, generate an image based on the transcription, or chat with GPT.",
# )

# # Gradio interface for ChatGPT-like functionality
# chat_iface = gr.Interface(
#     fn=chat_with_gpt,
#     inputs=gr.Textbox(label="Enter your prompt for ChatGPT"),
#     outputs=gr.Textbox(label="ChatGPT Response"),
#     title="ChatGPT",
#     description="Chat with GPT-like conversational AI.",
# )

# # Combined Gradio app
# iface = gr.TabbedInterface(
#     interface_list=[speech_to_text_iface, voice_to_image_and_chat_iface, chat_iface],
#     tab_names=["Speech-to-Text", "Voice-to-Image & Chat", "ChatGPT"]
# )

# # Launch Gradio interface
# iface.launch(debug=True, share=True)

import subprocess

# Install required libraries
subprocess.check_call(["pip", "install", "torch>=1.11.0"])
subprocess.check_call(["pip", "install", "transformers>=4.31.0"])
subprocess.check_call(["pip", "install", "diffusers>=0.14.0"])
subprocess.check_call(["pip", "install", "librosa"])
subprocess.check_call(["pip", "install", "accelerate>=0.20.1"])
subprocess.check_call(["pip", "install", "gradio>=3.35.2"])
subprocess.check_call(["pip", "install", "huggingface_hub"])

import os
import threading
import numpy as np
import librosa
import torch
import gradio as gr
from functools import lru_cache
from transformers import pipeline
from huggingface_hub import login
from diffusers import StableDiffusionPipeline, DPMSolverMultistepScheduler

# Ensure required dependencies are installed
def install_missing_packages():
    required_packages = {
        "librosa": None,
        "diffusers": ">=0.14.0",
        "gradio": ">=3.35.2",
        "huggingface_hub": None,
        "accelerate": ">=0.20.1",
        "transformers": ">=4.31.0"
    }
    for package, version in required_packages.items():
        try:
            __import__(package)
        except ImportError:
            package_name = f"{package}{version}" if version else package
            subprocess.check_call(["pip", "install", package_name])

install_missing_packages()

# Get Hugging Face token for authentication
hf_token = os.getenv("HF_TOKEN")
if hf_token:
    login(hf_token)
else:
    raise ValueError("HF_TOKEN environment variable not set.")

# Load speech-to-text model (Whisper)
speech_to_text = pipeline(
    "automatic-speech-recognition", 
    model="openai/whisper-tiny", 
    return_timestamps=True
)

# Load Stable Diffusion model for text-to-image
text_to_image = StableDiffusionPipeline.from_pretrained("runwayml/stable-diffusion-v1-5")
device = "cuda" if torch.cuda.is_available() else "cpu"
text_to_image.to(device)
text_to_image.enable_attention_slicing()
text_to_image.safety_checker = None
text_to_image.scheduler = DPMSolverMultistepScheduler.from_config(text_to_image.scheduler.config)

# Preprocess audio file into NumPy array
def preprocess_audio(audio_path):
    try:
        audio, sr = librosa.load(audio_path, sr=16000)  # Resample to 16kHz
        return np.array(audio, dtype=np.float32)
    except Exception as e:
        return f"Error in preprocessing audio: {str(e)}"

# Speech-to-text function with long-form transcription support
@lru_cache(maxsize=10)
def transcribe_audio(audio_path):
    try:
        audio_array = preprocess_audio(audio_path)
        if isinstance(audio_array, str):  # Error message from preprocessing
            return audio_array
        result = speech_to_text(audio_array)
        # Combine text from multiple segments for long-form transcription
        transcription = " ".join(segment["text"] for segment in result["chunks"])
        return transcription
    except Exception as e:
        return f"Error in transcription: {str(e)}"

# Text-to-image function
@lru_cache(maxsize=10)
def generate_image_from_text(text):
    try:
        image = text_to_image(text, height=256, width=256).images[0]  # Generate smaller images for speed
        return image
    except Exception as e:
        return f"Error in image generation: {str(e)}"

# Combined processing function
def process_audio_and_generate_results(audio_path):
    transcription_result = {"result": None}
    image_result = {"result": None}

    # Function to run transcription and image generation in parallel
    def transcription_thread():
        transcription_result["result"] = transcribe_audio(audio_path)

    def image_generation_thread():
        transcription = transcription_result["result"]
        if transcription and "Error" not in transcription:
            image_result["result"] = generate_image_from_text(transcription)

    # Start both tasks in parallel
    t1 = threading.Thread(target=transcription_thread)
    t2 = threading.Thread(target=image_generation_thread)

    t1.start()
    t2.start()

    t1.join()  # Wait for transcription to finish
    t2.join()  # Wait for image generation to finish

    transcription = transcription_result["result"]
    image = image_result["result"]

    if "Error" in transcription:
        return None, transcription
    if isinstance(image, str) and "Error" in image:
        return None, image

    return image, transcription

# Gradio interface for speech-to-text
speech_to_text_iface = gr.Interface(
    fn=transcribe_audio,
    inputs=gr.Audio(type="filepath", label="Upload audio file for transcription (WAV/MP3)"),
    outputs=gr.Textbox(label="Transcription"),
    title="Speech-to-Text Transcription",
    description="Upload an audio file to transcribe speech into text.",
)

# Gradio interface for voice-to-image
voice_to_image_iface = gr.Interface(
    fn=process_audio_and_generate_results,
    inputs=gr.Audio(type="filepath", label="Upload audio file (WAV/MP3)"),
    outputs=[gr.Image(label="Generated Image"), gr.Textbox(label="Transcription")],
    title="Voice-to-Image",
    description="Upload an audio file to transcribe speech to text and generate an image based on the transcription.",
)

# Combined Gradio app
iface = gr.TabbedInterface(
    interface_list=[speech_to_text_iface, voice_to_image_iface],
    tab_names=["Speech-to-Text", "Voice-to-Image"]
)

# Launch Gradio interface
iface.launch(debug=True, share=True)