app.py

from transformers import pipeline, BlipForConditionalGeneration, BlipProcessor, AutoTokenizer, AutoModelForSeq2SeqLM
import torchaudio
from torchaudio.transforms import Resample
import torch
import gradio as gr

# Initialize TTS model from Hugging Face
tts_model_name = "suno/bark"
tts = pipeline(task="text-to-speech", model=tts_model_name)

# Initialize Blip model for image captioning
model_id = "dblasko/blip-dalle3-img2prompt"
blip_model = BlipForConditionalGeneration.from_pretrained(model_id)
blip_processor = BlipProcessor.from_pretrained(model_id)

def generate_caption(image):
    # Generate caption from image using Blip model
    inputs = blip_processor(images=image, return_tensors="pt")
    pixel_values = inputs.pixel_values
    generated_ids = blip_model.generate(pixel_values=pixel_values, max_length=50)
    generated_caption = blip_processor.batch_decode(generated_ids, skip_special_tokens=True, temperature=0.8, top_k=40, top_p=0.9)[0]

    # Use TTS model to convert generated caption to audio
    audio_output = tts(generated_caption)
    audio_path = "generated_audio_resampled.wav"
    torchaudio.save(audio_path, torch.tensor(audio_output[0]), audio_output["sampling_rate"])

    return generated_caption, audio_path

# Create a Gradio interface with an image input, a textbox output, a button, and an audio player
demo = gr.Interface(
    fn=generate_caption,
    inputs=gr.Image(),
    outputs=[
        gr.Textbox(label="Generated caption"),
        gr.Button("Converts to Audio"),
        gr.Audio(type="filepath", label="Generated Audio")
    ],
    live=True
)
demo.launch(share=True)