app.py

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

# Initialize TTS model from TTS library
tts_model_path = "tts_models/multilingual/multi-dataset/xtts_v1"
tts = TTS(tts_model_path, gpu=True)

# 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
    tts.tts_to_file(text=generated_caption,
                    file_path="generated_audio.wav",
                    speaker_wav="/path/to/target/speaker.wav",
                    language="en")

    # Resample the audio to match the expected sampling rate
    waveform, sample_rate = torchaudio.load("generated_audio.wav")
    resampler = Resample(orig_freq=sample_rate, new_freq=24_000)
    waveform_resampled = resampler(waveform)

    # Save the resampled audio
    torchaudio.save("generated_audio_resampled.wav", waveform_resampled, 24_000)

    return generated_caption, "generated_audio_resampled.wav"

# 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("Convert to Audio"),
        gr.Audio(type="player", label="Generated Audio")
    ],
    live=True
)
demo.launch(share=True)