app.py

import gradio as gr
import whisper
from PIL import Image

import os
MY_SECRET_TOKEN=os.environ.get('HF_TOKEN_SD')

from diffusers import StableDiffusionPipeline

whisper_model = whisper.load_model("small")

device="cpu"

pipe = StableDiffusionPipeline.from_pretrained("CompVis/stable-diffusion-v1-4", use_auth_token=MY_SECRET_TOKEN)
pipe.to(device)

def get_transcribe(audio):
    audio = whisper.load_audio(audio)
    audio = whisper.pad_or_trim(audio)
    
    mel = whisper.log_mel_spectrogram(audio).to(whisper_model.device)
    
    #_, probs = whisper_model.detect_language(mel)
    
    options = whisper.DecodingOptions(task="translate", fp16 = False)
    result = whisper.decode(whisper_model, mel, options)
    
    print(result)
    print(result.text)
    return result.text

def get_images(audio): 
    prompt = get_transcribe(audio)
    #image = pipe(prompt, init_image=init_image)["sample"][0]
    images_list = pipe([prompt] * 2)
    images = []
    safe_image = Image.open(r"unsafe.png")
    for i, image in enumerate(images_list["sample"]):
        if(images_list["nsfw_content_detected"][i]):
            images.append(safe_image)
        else:
            images.append(image)
    
    return images
#inputs
audio = gr.Audio(label="Input Audio", show_label=False, source="microphone", type="filepath")
#outputs
gallery = gr.Gallery(label="Generated images", show_label=False, elem_id="gallery").style(grid=[2], height="auto")

gr.Interface(fn=get_images, inputs=audio, outputs=gallery).queue(max_size=10).launch(enable_queue=True)