import gradio as gr import pandas as pd from datasets import load_dataset from jiwer import wer, cer import os from datetime import datetime import re from huggingface_hub import login # Login to Hugging Face Hub (if token is available) token = os.environ.get("HG_TOKEN") if token: login(token) try: dataset = load_dataset("sudoping01/bambara-speech-recognition-benchmark", name="default")["eval"] references = {row["id"]: row["text"] for row in dataset} print(f"Loaded {len(references)} reference transcriptions") except Exception as e: print(f"Error loading dataset: {str(e)}") references = {} leaderboard_file = "leaderboard.csv" if not os.path.exists(leaderboard_file): sample_data = [ ["test_1", 0.2264, 0.1094, 0.1922, "2025-03-15 10:30:45"], ["test_2", 0.3264, 0.1094, 0.1922, "2025-03-15 10:30:45"], ] pd.DataFrame(sample_data, columns=["Model_Name", "WER", "CER", "Combined_Score", "timestamp"]).to_csv(leaderboard_file, index=False) print(f"Created new leaderboard file with sample data") else: leaderboard_df = pd.read_csv(leaderboard_file) if "Combined_Score" not in leaderboard_df.columns: leaderboard_df["Combined_Score"] = leaderboard_df["WER"] * 0.7 + leaderboard_df["CER"] * 0.3 leaderboard_df.to_csv(leaderboard_file, index=False) print(f"Added Combined_Score column to existing leaderboard") print(f"Loaded leaderboard with {len(leaderboard_df)} entries") def normalize_text(text): """Normalize text for WER/CER calculation""" if not isinstance(text, str): text = str(text) text = text.lower() text = re.sub(r'[^\w\s]', '', text) text = re.sub(r'\s+', ' ', text).strip() return text def calculate_metrics(predictions_df): """Calculate WER and CER for predictions.""" results = [] total_ref_words = 0 total_ref_chars = 0 for _, row in predictions_df.iterrows(): id_val = row["id"] if id_val not in references: continue reference = normalize_text(references[id_val]) hypothesis = normalize_text(row["text"]) if not reference or not hypothesis: continue reference_words = reference.split() hypothesis_words = hypothesis.split() reference_chars = list(reference) try: sample_wer = wer(reference, hypothesis) sample_cer = cer(reference, hypothesis) sample_wer = min(sample_wer, 2.0) sample_cer = min(sample_cer, 2.0) total_ref_words += len(reference_words) total_ref_chars += len(reference_chars) results.append({ "id": id_val, "reference": reference, "hypothesis": hypothesis, "ref_word_count": len(reference_words), "ref_char_count": len(reference_chars), "wer": sample_wer, "cer": sample_cer }) except Exception as e: print(f"Error processing sample {id_val}: {str(e)}") pass if not results: raise ValueError("No valid samples for WER/CER calculation") avg_wer = sum(item["wer"] for item in results) / len(results) avg_cer = sum(item["cer"] for item in results) / len(results) weighted_wer = sum(item["wer"] * item["ref_word_count"] for item in results) / total_ref_words weighted_cer = sum(item["cer"] * item["ref_char_count"] for item in results) / total_ref_chars return avg_wer, avg_cer, weighted_wer, weighted_cer, results def format_as_percentage(value): """Convert decimal to percentage with 2 decimal places""" return f"{value * 100:.2f}%" def prepare_leaderboard_for_display(df, sort_by="Combined_Score"): """Format leaderboard for display with ranking and percentages""" if df is None or len(df) == 0: return pd.DataFrame(columns=["Rank", "Model_Name", "WER (%)", "CER (%)", "Combined_Score (%)", "timestamp"]) display_df = df.copy() display_df = display_df.sort_values(sort_by) display_df.insert(0, "Rank", range(1, len(display_df) + 1)) for col in ["WER", "CER", "Combined_Score"]: if col in display_df.columns: display_df[f"{col} (%)"] = display_df[col].apply(lambda x: f"{x * 100:.2f}") return display_df def update_ranking(method): """Update leaderboard ranking based on selected method""" try: current_lb = pd.read_csv(leaderboard_file) if "Combined_Score" not in current_lb.columns: current_lb["Combined_Score"] = current_lb["WER"] * 0.7 + current_lb["CER"] * 0.3 sort_column = "Combined_Score" if method == "WER Only": sort_column = "WER" elif method == "CER Only": sort_column = "CER" return prepare_leaderboard_for_display(current_lb, sort_column) except Exception as e: print(f"Error updating ranking: {str(e)}") return pd.DataFrame(columns=["Rank", "Model_Name", "WER (%)", "CER (%)", "Combined_Score (%)", "timestamp"]) def process_submission(model_name, csv_file): """Process a new model submission""" if not model_name or not model_name.strip(): return "Error: Please provide a model name.", None if not csv_file: return "Error: Please upload a CSV file.", None try: df = pd.read_csv(csv_file) if len(df) == 0: return "Error: Uploaded CSV is empty.", None if set(df.columns) != {"id", "text"}: return f"Error: CSV must contain exactly 'id' and 'text' columns. Found: {', '.join(df.columns)}", None if df["id"].duplicated().any(): dup_ids = df[df["id"].duplicated()]["id"].unique() return f"Error: Duplicate IDs found: {', '.join(map(str, dup_ids[:5]))}", None missing_ids = set(references.keys()) - set(df["id"]) extra_ids = set(df["id"]) - set(references.keys()) if missing_ids: return f"Error: Missing {len(missing_ids)} IDs in submission. First few missing: {', '.join(map(str, list(missing_ids)[:5]))}", None if extra_ids: return f"Error: Found {len(extra_ids)} extra IDs not in reference dataset. First few extra: {', '.join(map(str, list(extra_ids)[:5]))}", None try: avg_wer, avg_cer, weighted_wer, weighted_cer, detailed_results = calculate_metrics(df) # Check for suspiciously low values if avg_wer < 0.001: return "Error: WER calculation yielded suspicious results (near-zero). Please check your submission CSV.", None except Exception as e: return f"Error calculating metrics: {str(e)}", None leaderboard = pd.read_csv(leaderboard_file) timestamp = datetime.now().strftime("%Y-%m-%d %H:%M:%S") combined_score = avg_wer * 0.7 + avg_cer * 0.3 if model_name in leaderboard["Model_Name"].values: idx = leaderboard[leaderboard["Model_Name"] == model_name].index leaderboard.loc[idx, "WER"] = avg_wer leaderboard.loc[idx, "CER"] = avg_cer leaderboard.loc[idx, "Combined_Score"] = combined_score leaderboard.loc[idx, "timestamp"] = timestamp updated_leaderboard = leaderboard else: new_entry = pd.DataFrame( [[model_name, avg_wer, avg_cer, combined_score, timestamp]], columns=["Model_Name", "WER", "CER", "Combined_Score", "timestamp"] ) updated_leaderboard = pd.concat([leaderboard, new_entry]) updated_leaderboard = updated_leaderboard.sort_values("Combined_Score") updated_leaderboard.to_csv(leaderboard_file, index=False) display_leaderboard = prepare_leaderboard_for_display(updated_leaderboard) return f"Submission processed successfully! WER: {format_as_percentage(avg_wer)}, CER: {format_as_percentage(avg_cer)}, Combined Score: {format_as_percentage(combined_score)}", display_leaderboard except Exception as e: return f"Error processing submission: {str(e)}", None def get_current_leaderboard(): """Get the current leaderboard data for display""" try: if os.path.exists(leaderboard_file): current_leaderboard = pd.read_csv(leaderboard_file) if "Combined_Score" not in current_leaderboard.columns: current_leaderboard["Combined_Score"] = current_leaderboard["WER"] * 0.7 + current_leaderboard["CER"] * 0.3 current_leaderboard.to_csv(leaderboard_file, index=False) return current_leaderboard else: return pd.DataFrame(columns=["Model_Name", "WER", "CER", "Combined_Score", "timestamp"]) except Exception as e: print(f"Error getting leaderboard: {str(e)}") return pd.DataFrame(columns=["Model_Name", "WER", "CER", "Combined_Score", "timestamp"]) def create_leaderboard_table(): """Create and format the leaderboard table for display""" leaderboard_data = get_current_leaderboard() return prepare_leaderboard_for_display(leaderboard_data) with gr.Blocks(title="Bambara ASR Leaderboard") as demo: gr.Markdown( """ # 🇲🇱 Bambara ASR Leaderboard This leaderboard tracks and evaluates speech recognition models for the Bambara language. Models are ranked based on Word Error Rate (WER), Character Error Rate (CER), and a combined score. ## Current Models Performance """ ) current_data = get_current_leaderboard() if len(current_data) > 0: best_model = current_data.sort_values("Combined_Score").iloc[0] gr.Markdown(f""" ### 🏆 Current Best Model: **{best_model['Model_Name']}** * WER: **{best_model['WER']*100:.2f}%** * CER: **{best_model['CER']*100:.2f}%** * Combined Score: **{best_model['Combined_Score']*100:.2f}%** """) with gr.Tabs() as tabs: with gr.TabItem("🏅 Model Rankings"): initial_leaderboard = create_leaderboard_table() ranking_method = gr.Radio( ["Combined Score (WER 70%, CER 30%)", "WER Only", "CER Only"], label="Ranking Method", value="Combined Score (WER 70%, CER 30%)" ) leaderboard_view = gr.DataFrame( value=initial_leaderboard, interactive=False, label="Models are ranked by selected metric - lower is better" ) ranking_method.change( fn=update_ranking, inputs=[ranking_method], outputs=[leaderboard_view] ) with gr.Accordion("Metrics Explanation", open=False): gr.Markdown( """ ## Understanding ASR Metrics ### Word Error Rate (WER) WER measures how accurately the ASR system recognizes whole words: * Lower values indicate better performance * Calculated as: (Substitutions + Insertions + Deletions) / Total Words * A WER of 0% means perfect transcription * A WER of 20% means approximately 1 in 5 words contains an error ### Character Error Rate (CER) CER measures accuracy at the character level: * More fine-grained than WER * Better at capturing partial word matches * Particularly useful for agglutinative languages like Bambara ### Combined Score * Weighted average: 70% WER + 30% CER * Provides a balanced evaluation of model performance * Used as the primary ranking metric """ ) with gr.TabItem("📊 Submit New Results"): gr.Markdown( """ ### Submit a new model for evaluation Upload a CSV file with the following format: * Must contain exactly two columns: 'id' and 'text' * The 'id' column should match the reference dataset IDs * The 'text' column should contain your model's transcriptions """ ) with gr.Row(): model_name_input = gr.Textbox( label="Model Name", placeholder="e.g., MALIBA-AI/bambara-asr" ) gr.Markdown("*Use a descriptive name to identify your model*") with gr.Row(): csv_upload = gr.File( label="Upload CSV File", file_types=[".csv"] ) gr.Markdown("*CSV with columns: id, text*") submit_btn = gr.Button("Submit", variant="primary") output_msg = gr.Textbox(label="Status", interactive=False) leaderboard_display = gr.DataFrame( label="Updated Leaderboard", value=initial_leaderboard, interactive=False ) submit_btn.click( fn=process_submission, inputs=[model_name_input, csv_upload], outputs=[output_msg, leaderboard_display] ) with gr.TabItem("📝 Benchmark Dataset"): gr.Markdown( """ ## About the Benchmark Dataset This leaderboard uses the **[sudoping01/bambara-speech-recognition-benchmark](https://huggingface.co/datasets/MALIBA-AI/bambara-speech-recognition-leaderboard)** dataset: * Contains diverse Bambara speech samples * Includes various speakers, accents, and dialects * Covers different speech styles and recording conditions * Transcribed and validated ### How to Generate Predictions To submit results to this leaderboard: 1. Download the audio files from the benchmark dataset 2. Run your ASR model on the audio files 3. Generate a CSV file with 'id' and 'text' columns 4. Submit your results using the form in the "Submit New Results" tab ### Evaluation Guidelines * Text is normalized (lowercase, punctuation removed) before metrics calculation * Extreme outliers are capped to prevent skewing results * All submissions are validated for format and completeness NB: This work is a collaboration between MALIBA-AI, RobotsMali AI4D-LAB and Djelia """ ) gr.Markdown( """ --- ### About MALIBA-AI **MALIBA-AI: Empowering Mali's Future Through Community-Driven AI Innovation** *"No Malian Language Left Behind"* This leaderboard is maintained by the MALIBA-AI initiative to track progress in Bambara speech recognition technology. For more information, visit [MALIBA-AI on Hugging Face](https://huggingface.co/MALIBA-AI). """ ) if __name__ == "__main__": demo.launch()