app.py

import gradio as gr # needs to be installed
from dataset import Dataset
from model import Model
from utils import compute_wer

# from utils import hf_login
# hf_login()

from huggingface_hub import login
import os

hf_token = os.getenv("HF_Token")
login(hf_token)


dataset = Dataset()
models = Model()


def run_tests (dataset_choice:str, model:str):

    MoDeL = Model()
    MoDeL.select(model)
    MoDeL.load()
    DaTaSeT = Dataset(100)
    DaTaSeT.load(dataset_choice)
    references, predictions = MoDeL.process(DaTaSeT)
    wer = compute_wer(references=references, predictions=predictions)
    return wer


def eval(data_subset:str, model_1:str, model_2:str)->str:

    wer_result_1 = run_tests(data_subset, model_1)
    wer_result_2 = run_tests(data_subset, model_2)

    return f"WER Score: {wer_result_1}", f"WER Score: {wer_result_2}"

def get_card(selected_model:str)->str:

    with open("cards.txt", "r") as f:
        cards = f.read()

    cards = cards.split("@@")
    for card in cards:
        if "ID: "+selected_model in card:
            return card

    return "Unknown Model"

def is_own(data_subset:str):
    if data_subset == "own":
        own_audio = gr.Audio(sources=['microphone'],streaming=False)
        own_transcription = gr.TextArea(lines=2)
        return own_audio, own_transcription
    own_audio = None
    own_transcription = None
    return own_audio, own_transcription

with gr.Blocks() as demo:


    gr.Markdown('# <p style="text-align: center;">VocalVenturer 💬</p>')
    gr.Markdown("-------")
    gr.Markdown("""#### Hello there, this is the VocalVenturer, this app is aimed at helping you making more informed model choices for ASR. Please choose a Data Subset to evalutate the Models on. You also have the opportunity to record and transcibe an own sample. The Models will be evaluated using the *WER* metric -> here is an amazing Leaderboard for it LINK""")

    with gr.Row():
        with gr.Column(scale=1):
            pass
        with gr.Column(scale=5):
            data_subset = gr.Radio(
                value="LibriSpeech Clean",
                choices=dataset.get_options(),
                label="Data subset / Own Sample",
            )
        with gr.Column(scale=1):
            pass

    with gr.Row():
        own_audio = gr.Audio(sources=['microphone'],streaming=False,visible=False)
        own_transcription = gr.TextArea(lines=2, visible=False)

        data_subset.change(is_own, inputs=[data_subset], outputs=[own_audio, own_transcription])


    with gr.Row():

        with gr.Column(scale=1):
            model_1 = gr.Dropdown(
                choices=models.get_options(),
                label="Select Model"
            )
            model_1_card = gr.Markdown("")

        with gr.Column(scale=1):
            model_2 = gr.Dropdown(
                choices=models.get_options(),
                label="Select Model"
            )
            model_2_card = gr.Markdown("")

        # with gr.Column(scale=1):
        #     model_3 = gr.Dropdown(
        #         choices=models.get_options(),
        #         label="Select Model"
        #     )
        #     model_3_card = gr.Markdown("")

        model_1.change(get_card, inputs=model_1, outputs=model_1_card)
        model_2.change(get_card, inputs=model_2, outputs=model_2_card)
        # model_3.change(get_card, inputs=model_3, outputs=model_3_card)

    eval_btn = gr.Button(
        value="Evaluate",
        variant="primary",
        size="sm"
    )
    

    
    gr.Markdown('## <p style="text-align: center;">Results</p>')
    gr.Markdown("-------")
    

    with gr.Row():

        with gr.Column(scale=1):
            results_model_1 = gr.Markdown("")

        with gr.Column(scale=1):
            results_model_2 = gr.Markdown("")

    eval_btn.click(eval, [data_subset, model_1, model_2], [results_model_1, results_model_2])

    

demo.launch(debug=True)