import gradio as gr
import numpy as np

import onnx_asr

models = {name: onnx_asr.load_model(name) for name in ["alphacep/vosk-model-ru", "alphacep/vosk-model-small-ru"]}


def recoginize(audio: tuple[int, np.ndarray]):
    sample_rate, waveform = audio
    waveform = waveform.astype(np.float32) / 2 ** (8 * waveform.itemsize - 1)
    return [[name, model.recognize(waveform, sample_rate=sample_rate)] for name, model in models.items()]


demo = gr.Interface(
    fn=recoginize,
    inputs=[gr.Audio(min_length=1, max_length=10)],
    outputs=[gr.Dataframe(headers=["Model", "result"], wrap=True, show_fullscreen_button=True)],
    flagging_mode="never",
)

demo.launch()