app.py

import gradio as gr
import pandas as pd

from src.about import (
    CITATION_BUTTON_LABEL,
    CITATION_BUTTON_TEXT,
    INTRODUCTION_TEXT,
    ABOUT_TEXT,
    TITLE,
)
from src.display.css_html_js import custom_css


def load_results():
    """Load and process results from CSV file"""
    try:
        df = pd.read_csv("results.csv")
        
        # Get WER by dataset for each model
        wer_by_dataset = df.pivot_table(
            index='model_id', 
            columns='dataset', 
            values='wer', 
            aggfunc='mean'
        ).round(2)
        
        # Calculate overall average WER
        wer_by_dataset['Average WER'] = df.groupby('model_id')['wer'].mean().round(2)
        
        # Calculate RTFx properly: sum(total_audio_length) / sum(total_time)
        audio_time_sums = df.groupby('model_id').agg({
            'total_audio_length': 'sum',
            'total_time': 'sum'
        })
        rtfx_calculated = (audio_time_sums['total_audio_length'] / audio_time_sums['total_time']).round(2)
        
        # Combine all metrics
        model_stats = wer_by_dataset.copy()
        model_stats['RTFx'] = rtfx_calculated
        
        # Set RTFx to NA for ElevenLabs (API-based, not local model)
        elevenlabs_mask = model_stats.index.str.contains('elevenlabs', case=False, na=False)
        model_stats.loc[elevenlabs_mask, 'RTFx'] = 'N/A'
        
        # Sort by average WER (lower is better)
        model_stats = model_stats.sort_values('Average WER')
        
        # Reset index to make model_id a column
        model_stats = model_stats.reset_index()
        
        # Reorder columns: Model, Average WER first, then Datarisas, then other datasets, then RTFx
        dataset_columns = [col for col in model_stats.columns if col not in ['model_id', 'Average WER', 'RTFx']]
        
        # Put datarisas first, then other datasets
        datarisas_col = [col for col in dataset_columns if 'datarisas' in col.lower()]
        other_dataset_cols = [col for col in dataset_columns if 'datarisas' not in col.lower()]
        ordered_dataset_cols = datarisas_col + other_dataset_cols
        
        new_column_order = ['model_id', 'Average WER'] + ordered_dataset_cols + ['RTFx']
        model_stats = model_stats[new_column_order]
        
        # Convert model names to appropriate links
        def create_model_link(model_name):
            if 'elevenlabs' in model_name.lower():
                return f'<a href="https://elevenlabs.io/speech-to-text" target="_blank">{model_name}</a>'
            else:
                return f'<a href="https://huggingface.co/{model_name}" target="_blank">{model_name}</a>'
        
        model_stats['model_id'] = model_stats['model_id'].apply(create_model_link)
        
        # Rename columns for better display
        column_mapping = {'model_id': 'Model', 'Average WER': 'Average WER ⬇️', 'RTFx': 'RTFx ⬆️'}
        # Add arrows to dataset WER columns
        for col in dataset_columns:
            column_mapping[col] = f'{col.replace("_", " ").title()} WER ⬇️'
        
        model_stats = model_stats.rename(columns=column_mapping)
        
        return model_stats
    
    except FileNotFoundError:
        # Return empty dataframe if CSV doesn't exist
        return pd.DataFrame(columns=['Model', 'Average WER ⬇️', 'RTFx ⬆️'])


# Load results
leaderboard_df = load_results()

demo = gr.Blocks(css=custom_css)
with demo:
    gr.HTML(TITLE)
    gr.Markdown(INTRODUCTION_TEXT, elem_classes="markdown-text")

    with gr.Tabs(elem_classes="tab-buttons") as tabs:
        with gr.TabItem("🏅 Chilean Spanish ASR Leaderboard", elem_id="leaderboard-tab", id=0):
            gr.Dataframe(
                value=leaderboard_df,
                interactive=False,
                wrap=True,
                datatype=["markdown"] + ["number"] * (len(leaderboard_df.columns) - 1)
            )

        with gr.TabItem("📝 About", elem_id="about-tab", id=1):
            gr.Markdown(ABOUT_TEXT, elem_classes="markdown-text")

    with gr.Row():
        with gr.Accordion("📙 Citation", open=False):
            citation_button = gr.Textbox(
                value=CITATION_BUTTON_TEXT,
                label=CITATION_BUTTON_LABEL,
                lines=20,
                elem_id="citation-button",
                show_copy_button=True,
            )

demo.launch()