app.py

import gradio as gr
from huggingsound import SpeechRecognitionModel
from diffusers import StableDiffusionPipeline, DPMSolverMultistepScheduler
from pydub import AudioSegment
import io

# Función para convertir la tasa de muestreo del audio de entrada
def convert_sampling_rate(audio, target_sr=48000):
    if isinstance(audio, tuple):
        audio_data, sample_rate = audio
        audio_data = audio_data.to_bytes((audio_data.bit_length() + 7) // 8, 'big')
    else:
        audio_data = audio
    audio = AudioSegment.from_wav(io.BytesIO(audio_data))
    audio = audio.set_frame_rate(target_sr)
    return audio.raw_dataa

def modelo1(audio):
    # Convertir la tasa de muestreo del audio
    audio = convert_sampling_rate(audio)
    
    model = SpeechRecognitionModel("jonatasgrosman/wav2vec2-large-xlsr-53-english")
    transcriptions = model.transcribe(audio)
    return transcriptions

def modelo2(text):
    model_id = "stabilityai/stable-diffusion-2-1"

    # Use the DPMSolverMultistepScheduler (DPM-Solver++) scheduler here instead
    pipe = StableDiffusionPipeline.from_pretrained(model_id, torch_dtype=torch.float16)
    pipe.scheduler = DPMSolverMultistepScheduler.from_config(pipe.scheduler.config)
    pipe = pipe.to("cuda")

    image = pipe(text).images[0]
    return image

def execution(audio):
    modelo1res = modelo1(audio)
    modelo2res = modelo2(modelo1res)
    return modelo2res

if __name__ == "__main__":
    demo = gr.Interface(fn=execution, inputs="audio", outputs="image")
    demo.launch()