app.py

import gradio as gr
import whisper
from transformers import MBartForConditionalGeneration, MBart50TokenizerFast

# función para transcribir el audio

model = MBartForConditionalGeneration.from_pretrained("SnypzZz/Llama2-13b-Language-translate")
tokenizer = MBart50TokenizerFast.from_pretrained("SnypzZz/Llama2-13b-Language-translate", src_lang="en_XX")

dropdown = gr.Dropdown(["de_DE", "es_XX", "fr_XX", "sv_SE", "ru_RU"], label="Choose Output Language")

def execute(input, dropdown_value):
    model_inputs = tokenizer(input, return_tensors="pt")
    generated_tokens = model.generate(
        **model_inputs,
        forced_bos_token_id=tokenizer.lang_code_to_id[dropdown_value]
    )
    output = tokenizer.batch_decode(generated_tokens, skip_special_tokens=True)[0]
    output = output.strip("[]' ") 
    return output

def whisper_transcript(model_size, audio_file):
    
    source = audio_file
    loaded_model = whisper.load_model(model_size)
    transcript = loaded_model.transcribe(source, language="english")

    return transcript["text"]


# interfaz gradio
gradio_ui = gr.Interface(
    fn=whisper_transcript,
    theme="Nymbo/Nymbo_Theme",
    title="Transcribir audios en inglés a texto",
    description="**Cómo usar**: Elegir uno de los 4 modelos, subir un audio o grabarlo y clicar el botón de Submit.",
    article="**Nota**: Exclusivo para audios en inglés.",
    inputs=[
        gr.Dropdown(
            label="Select Model",
            choices=[
                "tiny.en",
                "base.en",
                "small.en",
                "medium.en",
            ],
            value="base",
        ),
        gr.Audio(label="Upload Audio File", sources=["upload", "microphone"], type="filepath"),
    ],
    outputs=gr.Textbox(label="Whisper Transcript"),
)

gradio_ui.queue().launch()