Upload folder using huggingface_hub

content.py

@@ -0,0 +1,169 @@
+
+displayname2datasetname = {
+    'LibriSpeech-Clean'    : 'librispeech_test_clean',
+    'LibriSpeech-Other'    : 'librispeech_test_other',
+    'CommonVoice-15-EN'    : 'common_voice_15_en_test',
+    'Peoples-Speech'       : 'peoples_speech_test',
+    'GigaSpeech-1'         : 'gigaspeech_test',
+    'Earnings-21'          : 'earnings21_test',
+    'Earnings-22'          : 'earnings22_test',
+    'TED-LIUM-3'           : 'tedlium3_test',
+    'TED-LIUM-3-LongForm'  : 'tedlium3_long_form_test',
+    'AISHELL-ASR-ZH'       : 'aishell_asr_zh_test',
+    'CoVoST2-EN-ID'        : 'covost2_en_id_test',
+    'CoVoST2-EN-ZH'        : 'covost2_en_zh_test',
+    'CoVoST2-EN-TA'        : 'covost2_en_ta_test',
+    'CoVoST2-ID-EN'        : 'covost2_id_en_test',
+    'CoVoST2-ZH-EN'        : 'covost2_zh_en_test',
+    'CoVoST2-TA-EN'        : 'covost2_ta_en_test',
+    'CN-College-Listen-MCQ': 'cn_college_listen_mcq_test',
+    'DREAM-TTS-MCQ'        : 'dream_tts_mcq_test',
+    'SLUE-P2-SQA5'         : 'slue_p2_sqa5_test',
+    'Public-SG-Speech-QA'  : 'public_sg_speech_qa_test',
+    'Spoken-SQuAD'         : 'spoken_squad_test',
+    'OpenHermes-Audio'     : 'openhermes_audio_test',
+    'ALPACA-Audio'         : 'alpaca_audio_test',
+    'WavCaps'              : 'wavcaps_test',
+    'AudioCaps'            : 'audiocaps_test',
+    'Clotho-AQA'           : 'clotho_aqa_test',
+    'WavCaps-QA'           : 'wavcaps_qa_test',
+    'AudioCaps-QA'         : 'audiocaps_qa_test',
+    'VoxCeleb-Accent'      : 'voxceleb_accent_test',
+    'MNSC-AR-Sentence'     : 'imda_ar_sentence',
+    'MNSC-AR-Dialogue'     : 'imda_ar_dialogue',
+    'VoxCeleb-Gender'      : 'voxceleb_gender_test',
+    'IEMOCAP-Gender'       : 'iemocap_gender_test',
+    'IEMOCAP-Emotion'      : 'iemocap_emotion_test',
+    'MELD-Sentiment'       : 'meld_sentiment_test',
+    'MELD-Emotion'         : 'meld_emotion_test',
+    'MuChoMusic'           : 'muchomusic_test',
+    'MNSC-PART1-ASR'       : 'imda_part1_asr_test',
+    'MNSC-PART2-ASR'       : 'imda_part2_asr_test',
+    'MNSC-PART3-ASR'       : 'imda_part3_30s_asr_test',
+    'MNSC-PART4-ASR'       : 'imda_part4_30s_asr_test',
+    'MNSC-PART5-ASR'       : 'imda_part5_30s_asr_test',
+    'MNSC-PART6-ASR'       : 'imda_part6_30s_asr_test',
+    'MNSC-PART3-SQA'       : 'imda_part3_30s_sqa_human_test',
+    'MNSC-PART4-SQA'       : 'imda_part4_30s_sqa_human_test',
+    'MNSC-PART5-SQA'       : 'imda_part5_30s_sqa_human_test',
+    'MNSC-PART6-SQA'       : 'imda_part6_30s_sqa_human_test',
+    'MNSC-PART3-SDS'       : 'imda_part3_30s_ds_human_test',
+    'MNSC-PART4-SDS'       : 'imda_part4_30s_ds_human_test',
+    'MNSC-PART5-SDS'       : 'imda_part5_30s_ds_human_test',
+    'MNSC-PART6-SDS'       : 'imda_part6_30s_ds_human_test',
+
+    'CNA'             : 'cna_test',
+    'IDPC'            : 'idpc_test',
+    'Parliament'      : 'parliament_test',
+    'UKUS-News'       : 'ukusnews_test',
+    'Mediacorp'       : 'mediacorp_test',
+    'IDPC-Short'      : 'idpc_short_test',
+    'Parliament-Short': 'parliament_short_test',
+    'UKUS-News-Short' : 'ukusnews_short_test',
+    'Mediacorp-Short' : 'mediacorp_short_test',
+    'YouTube ASR: English with Singapore Content': 'ytb_asr_batch1',
+    'YouTube ASR: English with Strong Emotion': 'ytb_asr_batch2',
+    'YouTube ASR: Malay with English Prompt': 'ytb_asr_batch3_ms',
+    'YouTube ASR: Malay with Malay Prompt': 'ytb_asr_batch3_ms_ms_prompt',
+    
+    'SEAME-Dev-Mandarin'   : 'seame_dev_man',
+    'SEAME-Dev-Singlish'   : 'seame_dev_sge',
+
+    'YouTube SQA: English with Singapore Content': 'ytb_sqa_batch1',
+    'YouTube SDS: English with Singapore Content': 'ytb_sds_batch1',
+    'YouTube PQA: English with Singapore Content': 'ytb_pqa_batch1',
+
+}
+
+datasetname2diaplayname = {datasetname: displayname for displayname, datasetname in displayname2datasetname.items()}
+
+
+dataset_diaplay_information = {
+    'LibriSpeech-Clean'    : 'A clean, high-quality testset of the LibriSpeech dataset, used for ASR testing.',
+    'LibriSpeech-Other'    : 'A more challenging, noisier testset of the LibriSpeech dataset for ASR testing.',
+    'CommonVoice-15-EN'    : 'Test set from the Common Voice project, which is a crowd-sourced, multilingual speech dataset.',
+    'Peoples-Speech'       : 'A large-scale, open-source speech recognition dataset, with diverse accents and domains.',
+    'GigaSpeech-1'         : 'A large-scale ASR dataset with diverse audio sources like podcasts, interviews, etc.',
+    'Earnings-21'          : 'ASR test dataset focused on earnings calls from 2021, with professional speech and financial jargon.',
+    'Earnings-22'          : 'Similar to Earnings21, but covering earnings calls from 2022.',
+    'TED-LIUM-3'           : 'A test set derived from TED talks, covering diverse speakers and topics.',
+    'TED-LIUM-3-LongForm'  : 'A longer version of the TED-LIUM dataset, containing extended audio samples. This poses challenges to existing fusion methods in handling long audios. However, it provides benchmark for future development.',
+    'AISHELL-ASR-ZH'       : 'ASR test dataset for Mandarin Chinese, based on the Aishell dataset.',
+    'CoVoST2-EN-ID'        : 'CoVoST 2 dataset for speech translation from English to Indonesian.',
+    'CoVoST2-EN-ZH'        : 'CoVoST 2 dataset for speech translation from English to Chinese.',
+    'CoVoST2-EN-TA'        : 'CoVoST 2 dataset for speech translation from English to Tamil.',
+    'CoVoST2-ID-EN'        : 'CoVoST 2 dataset for speech translation from Indonesian to English.',
+    'CoVoST2-ZH-EN'        : 'CoVoST 2 dataset for speech translation from Chinese to English.',
+    'CoVoST2-TA-EN'        : 'CoVoST 2 dataset for speech translation from Tamil to English.',
+    'CN-College-Listen-MCQ': 'Chinese College English Listening Test, with multiple-choice questions.',
+    'DREAM-TTS-MCQ'        : 'DREAM dataset for spoken question-answering, derived from textual data and synthesized speech.',
+    'SLUE-P2-SQA5'         : 'Spoken Language Understanding Evaluation (SLUE) dataset, part 2, focused on QA tasks.',
+    'Public-SG-Speech-QA'  : 'Public dataset for speech-based question answering, gathered from Singapore.',
+    'Spoken-SQuAD'         : 'Spoken SQuAD dataset, based on the textual SQuAD dataset, converted into audio.',
+    'OpenHermes-Audio'     : 'Test set for spoken instructions. Synthesized from the OpenHermes dataset.',
+    'ALPACA-Audio'         : 'Spoken version of the ALPACA dataset, used for evaluating instruction following in audio.',
+    'WavCaps'              : 'WavCaps is a dataset for testing audio captioning, where models generate textual descriptions of audio clips.',
+    'AudioCaps'            : 'AudioCaps dataset, used for generating captions from general audio events.',
+    'Clotho-AQA'           : 'Clotho dataset adapted for audio-based question answering, containing audio clips and questions.',
+    'WavCaps-QA'           : 'Question-answering test dataset derived from WavCaps, focusing on audio content.',
+    'AudioCaps-QA'         : 'AudioCaps adapted for question-answering tasks, using audio events as input for Q&A.',
+    'VoxCeleb-Accent'      : 'Test dataset for accent recognition, based on VoxCeleb, a large speaker identification dataset.',
+    'MNSC-AR-Sentence'     : 'Accent recognition based on the IMDA NSC dataset, focusing on sentence-level accents.',
+    'MNSC-AR-Dialogue'     : 'Accent recognition based on the IMDA NSC dataset, focusing on dialogue-level accents.',
+
+    'VoxCeleb-Gender': 'Test dataset for gender classification, also derived from VoxCeleb.',
+    'IEMOCAP-Gender' : 'Gender classification based on the IEMOCAP dataset.',
+    'IEMOCAP-Emotion': 'Emotion recognition test data from the IEMOCAP dataset, focusing on identifying emotions in speech.',
+    'MELD-Sentiment' : 'Sentiment recognition from speech using the MELD dataset, classifying positive, negative, or neutral sentiments.',
+    'MELD-Emotion'   : 'Emotion classification in speech using MELD, detecting specific emotions like happiness, anger, etc.',
+    'MuChoMusic'     : 'Test dataset for music understanding, from paper: MuChoMusic: Evaluating Music Understanding in Multimodal Audio-Language Models.',
+    'MNSC-PART1-ASR' : 'Speech recognition test data from the IMDA NSC project, Part 1.',
+    'MNSC-PART2-ASR' : 'Speech recognition test data from the IMDA NSC project, Part 2.',
+    'MNSC-PART3-ASR' : 'Speech recognition test data from the IMDA NSC project, Part 3.',
+    'MNSC-PART4-ASR' : 'Speech recognition test data from the IMDA NSC project, Part 4.',
+    'MNSC-PART5-ASR' : 'Speech recognition test data from the IMDA NSC project, Part 5.',
+    'MNSC-PART6-ASR' : 'Speech recognition test data from the IMDA NSC project, Part 6.',
+    'MNSC-PART3-SQA' : 'Multitak National Speech Corpus (MNSC) dataset, Question answering task, Part 3.',
+    'MNSC-PART4-SQA' : 'Multitak National Speech Corpus (MNSC) dataset, Question answering task, Part 4.',
+    'MNSC-PART5-SQA' : 'Multitak National Speech Corpus (MNSC) dataset, Question answering task, Part 5.',
+    'MNSC-PART6-SQA' : 'Multitak National Speech Corpus (MNSC) dataset, Question answering task, Part 6.',
+    'MNSC-PART3-SDS' : 'Multitak National Speech Corpus (MNSC) dataset, dialogue summarization task, Part 3.',
+    'MNSC-PART4-SDS' : 'Multitak National Speech Corpus (MNSC) dataset, dialogue summarization task, Part 4.',
+    'MNSC-PART5-SDS' : 'Multitak National Speech Corpus (MNSC) dataset, dialogue summarization task, Part 5.',
+    'MNSC-PART6-SDS' : 'Multitak National Speech Corpus (MNSC) dataset, dialogue summarization task, Part 6.',
+
+    'CNA'             : 'Under Development',
+    'IDPC'            : 'Under Development',
+    'Parliament'      : 'Under Development',
+    'UKUS-News'       : 'Under Development',
+    'Mediacorp'       : 'Under Development',
+    'IDPC-Short'      : 'Under Development',
+    'Parliament-Short': 'Under Development',
+    'UKUS-News-Short' : 'Under Development',
+    'Mediacorp-Short' : 'Under Development',
+    'YouTube ASR: English Singapore Content'  : '''\nYouTube Evaluation Dataset for ASR Task: This dataset include English and Singlish with Singapore Content.''',
+    'YouTube ASR: English with Strong Emotion'  : '\nYouTube Evaluation Dataset for ASR Task. English with strong emotions',
+    'YouTube ASR: Malay English Prompt': 'YouTube ASR Dataset, Malay and Malay-English CondeSwitch',
+    'YouTube ASR: Malay with Malay Prompt': 'YouTube ASR Dataset, Malay and Malay-English CondeSwitch. Use Malay prompts',
+    
+    'SEAME-Dev-Mandarin'   : 'Under Development',
+    'SEAME-Dev-Singlish'   : 'Under Development',
+
+    'YouTube SQA: English with Singapore Content': 'Under Development',
+    'YouTube SDS: English with Singapore Content': 'Under Development',
+    'YouTube PQA: English with Singapore Content': 'Under Development',
+    
+    
+                }
+
+
+
+
+metrics_info = {
+    'wer'                    : 'Word Error Rate (WER) - The Lower, the better.',
+    'llama3_70b_judge_binary': 'Model-as-a-Judge Peformance. Using LLAMA-3-70B. Scale from 0-100. The higher, the better.',
+    'llama3_70b_judge'       : 'Model-as-a-Judge Peformance. Using LLAMA-3-70B. Scale from 0-100. The higher, the better.',
+    'meteor'                 : 'METEOR Score. The higher, the better.',
+    'bleu'                   : 'BLEU Score. The higher, the better.',
+}
+

draw_diagram.py

@@ -0,0 +1,223 @@
+import streamlit as st
+import pandas as pd
+import numpy as np
+from streamlit_echarts import st_echarts
+from app.show_examples import *
+from app.content import *
+
+import pandas as pd
+
+from model_information import get_dataframe
+info_df = get_dataframe()
+
+
+def draw(folder_name, category_name, displayname, metrics, cus_sort=True):
+    
+    folder = f"./results_organized/{metrics}/"
+
+    # Load the results from CSV
+    data_path = f'{folder}/{category_name.lower()}.csv'
+    chart_data = pd.read_csv(data_path).round(3)
+    
+    dataset_name = displayname2datasetname[displayname]
+    chart_data = chart_data[['Model', dataset_name]]
+
+    # Rename to proper display name
+    chart_data = chart_data.rename(columns=datasetname2diaplayname)
+
+    st.markdown("""
+                <style>
+                .stMultiSelect [data-baseweb=select] span {
+                    max-width: 800px;
+                    font-size: 0.9rem;
+                    background-color: #3C6478 !important; /* Background color for selected items */
+                    color: white; /* Change text color */
+                    back
+                }
+                </style>
+                """, unsafe_allow_html=True)
+    
+    # remap model names
+    display_model_names = {key.strip() :val.strip() for key, val in zip(info_df['Original Name'], info_df['Proper Display Name'])}
+    chart_data['model_show'] = chart_data['Model'].map(lambda x: display_model_names.get(x, x))
+
+
+    models = st.multiselect("Please choose the model", 
+                            sorted(chart_data['model_show'].tolist()), 
+                            default = sorted(chart_data['model_show'].tolist()),
+                            )
+    
+    chart_data = chart_data[chart_data['model_show'].isin(models)]
+    chart_data = chart_data.sort_values(by=[displayname], ascending=cus_sort).dropna(axis=0)
+
+    if len(chart_data) == 0: return
+    
+    
+    # = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = =
+    '''
+    Show Table
+    '''
+    with st.container():
+        st.markdown('##### TABLE')
+
+        
+        model_link = {key.strip(): val for key, val in zip(info_df['Proper Display Name'], info_df['Link'])}
+
+        chart_data['model_link'] = chart_data['model_show'].map(model_link) 
+
+        chart_data_table = chart_data[['model_show', chart_data.columns[1], chart_data.columns[3]]]
+
+        # Format numeric columns to 2 decimal places
+        #chart_data_table[chart_data_table.columns[1]] = chart_data_table[chart_data_table.columns[1]].apply(lambda x: round(float(x), 3) if isinstance(float(x), (int, float)) else float(x))
+        cur_dataset_name = chart_data_table.columns[1]
+
+
+        def highlight_first_element(x):
+                # Create a DataFrame with the same shape as the input
+                df_style = pd.DataFrame('', index=x.index, columns=x.columns)
+                # Apply background color to the first element in row 0 (df[0][0])
+                # df_style.iloc[0, 1] = 'background-color: #b0c1d7; color: white'
+                df_style.iloc[0, 1] = 'background-color: #b0c1d7'
+                
+                return df_style
+
+        if cur_dataset_name in [
+                            'LibriSpeech-Clean',
+                            'LibriSpeech-Other',
+                            'CommonVoice-15-EN',
+                            'Peoples-Speech',
+                            'GigaSpeech-1',
+                            'Earnings-21',
+                            'Earnings-22',
+                            'TED-LIUM-3',
+                            'TED-LIUM-3-LongForm',
+                            'AISHELL-ASR-ZH',
+                            'MNSC-PART1-ASR',
+                            'MNSC-PART2-ASR',
+                            'MNSC-PART3-ASR',
+                            'MNSC-PART4-ASR',
+                            'MNSC-PART5-ASR',
+                            'MNSC-PART6-ASR',
+                            'CNA',
+                            'IDPC',
+                            'Parliament',
+                            'UKUS-News',
+                            'Mediacorp',
+                            'IDPC-Short',
+                            'Parliament-Short',
+                            'UKUS-News-Short',
+                            'Mediacorp-Short',
+                            'YTB-ASR-Batch1',
+                            'YTB-ASR-Batch2',
+                            'SEAME-Dev-Man',
+                            'SEAME-Dev-Sge',
+                            ]:
+            
+            chart_data_table = chart_data_table.sort_values(
+                    by=chart_data_table.columns[1],
+                    ascending=True
+                ).reset_index(drop=True)
+        else:
+            chart_data_table = chart_data_table.sort_values(
+                    by=chart_data_table.columns[1],
+                    ascending=False
+                ).reset_index(drop=True)
+            
+
+        styled_df = chart_data_table.style.format(
+            {chart_data_table.columns[1]: "{:.3f}"}
+        ).apply(
+            highlight_first_element, axis=None
+        )
+
+
+        st.dataframe(
+                styled_df,
+                column_config={
+                    'model_show': 'Model',
+                    chart_data_table.columns[1]: {'alignment': 'left'},
+                    "model_link": st.column_config.LinkColumn(
+                        "Model Link",
+                    ),
+                },
+                hide_index=True,
+                use_container_width=True
+            )
+        
+
+    # = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = =
+    '''
+    Show Chart
+    '''
+
+    # Initialize a session state variable for toggling the chart visibility
+    if "show_chart" not in st.session_state:
+        st.session_state.show_chart = False
+
+    # Create a button to toggle visibility
+    if st.button("Show Chart"):
+        st.session_state.show_chart = not st.session_state.show_chart
+
+    if st.session_state.show_chart:
+
+        with st.container():
+            st.markdown('##### CHART')
+
+            # Get Values
+            data_values = chart_data.iloc[:, 1]
+            
+            # Calculate Q1 and Q3
+            q1 = data_values.quantile(0.25)
+            q3 = data_values.quantile(0.75)
+
+            # Calculate IQR
+            iqr = q3 - q1
+
+            # Define lower and upper bounds (1.5*IQR is a common threshold)
+            lower_bound = q1 - 1.5 * iqr
+            upper_bound = q3 + 1.5 * iqr
+
+            # Filter data within the bounds
+            filtered_data = data_values[(data_values >= lower_bound) & (data_values <= upper_bound)]
+
+            # Calculate min and max values after outlier handling
+            min_value = round(filtered_data.min() - 0.1 * filtered_data.min(), 3)
+            max_value = round(filtered_data.max() + 0.1 * filtered_data.max(), 3)
+
+            options = {
+                # "title": {"text": f"{dataset_name}"},
+                "tooltip": {
+                    "trigger": "axis",
+                    "axisPointer": {"type": "cross", "label": {"backgroundColor": "#6a7985"}},
+                    "triggerOn": 'mousemove',
+                },
+                "legend": {"data": ['Overall Accuracy']},
+                "toolbox": {"feature": {"saveAsImage": {}}},
+                "grid": {"left": "3%", "right": "4%", "bottom": "3%", "containLabel": True},
+                "xAxis": [
+                    {
+                        "type": "category",
+                        "boundaryGap": True,
+                        "triggerEvent": True,
+                        "data":  chart_data['model_show'].tolist(),
+                    }
+                ],
+                "yAxis": [{"type": "value", 
+                            "min": min_value,
+                            "max": max_value, 
+                            "boundaryGap": True
+                            # "splitNumber": 10
+                            }],
+                "series": [{
+                        "name": f"{dataset_name}",
+                        "type": "bar",
+                        "data": chart_data[f'{displayname}'].tolist(),
+                    }],
+            }
+            
+            events = {
+                "click": "function(params) { return params.value }"
+            }
+
+            value = st_echarts(options=options, events=events, height="500px")
+            

pages.py

@@ -0,0 +1,626 @@
+import streamlit as st
+from app.draw_diagram import *
+from app.content import *
+from app.summarization import *
+from app.show_examples import *
+
+def dataset_contents(dataset, metrics):
+    
+    custom_css = """
+                <style>
+                .my-dataset-info {
+                # background-color: #F9EBEA;
+                # padding: 10px;
+                color: #050505;
+                font-style: normal;
+                font-size: 8px;
+                height: auto;
+                }
+                </style>
+                """
+    st.markdown(custom_css, unsafe_allow_html=True)
+    st.markdown(f"""<div class="my-dataset-info">
+                    <p><b>About this dataset</b>: {dataset}</p>
+                    </div>""", unsafe_allow_html=True)
+    st.markdown(f"""<div class="my-dataset-info">
+                    <p><b>About this metric</b>: {metrics}</p>
+                    </div>""", unsafe_allow_html=True)
+
+
+def dashboard():
+
+    with st.container():
+        st.title("Leaderboard for AudioBench")
+   
+        st.markdown("""
+            [gh1]: https://github.com/AudioLLMs/AudioBench
+            [gh2]: https://github.com/AudioLLMs/AudioBench
+            **Toolkit:** [![GitHub Repo stars](https://img.shields.io/github/stars/AudioLLMs/AudioBench?style=social)][gh1] | 
+            [**Paper @ NAACL 2025**](https://arxiv.org/abs/2406.16020) | 
+            **Resource for AudioLLMs:** [![GitHub Repo stars](https://img.shields.io/github/stars/AudioLLMs/Awesome-Audio-LLM?style=social)][gh2]
+            """)
+
+
+    st.markdown("""
+            #### Recent updates
+            - **Jan. 2025**: AudioBench is officially accepted to NAACL 2025!
+            - **Jan. 2025**: Update the layout.
+            - **Dec. 2024**: Added MuChoMusic dataset for Music Understanding - MCQ Questions. From Paper: https://arxiv.org/abs/2408.01337.
+            - **Dec. 2024**: Singlish ASR task added! The datasets are available on [HF](https://huggingface.co/datasets/MERaLiON/MNSC).
+            - **Dec. 2024**: Updated layout and added support for comparison between models with similar sizes. 1) Reorganized layout for a better user experience. 2) Added performance summary for each task.
+            - **Aug. 2024**: Initial leaderboard is now online.
+            """)
+
+    st.divider()
+    
+    st.markdown("""
+                #### Evaluating Audio-based Large Language Models
+                
+                - AudioBench is a comprehensive evaluation benchmark designed for general instruction-following audio large language models.
+                - AudioBench is an evaluation benchmark that we continually improve and maintain.
+                
+                Below are the initial 26 datasets that are included in AudioBench. We are now exteneded to over 40 datasets and going to extend to more in the future.
+                """
+                )
+
+
+    with st.container():
+        
+        st.markdown('''
+                ''')
+        
+        st.markdown("###### :dart: Our Benchmark includes: ")
+        cols = st.columns(8)
+        cols[0].metric(label="Tasks", value=">8")
+        cols[1].metric(label="Datasets", value=">40")
+        cols[2].metric(label="Evaluated Models", value=">5")
+    
+    st.divider()
+    with st.container():
+        left_co, right_co = st.columns([1, 0.1])
+
+        with left_co:
+            st.markdown("""
+                        ##### Citations :round_pushpin:
+                        ```
+                        @article{wang2024audiobench,
+                            title={AudioBench: A Universal Benchmark for Audio Large Language Models},
+                            author={Wang, Bin and Zou, Xunlong and Lin, Geyu and Sun, Shuo and Liu, Zhuohan and Zhang, Wenyu and Liu, Zhengyuan and Aw, AiTi and Chen, Nancy F},
+                            journal={NAACL},
+                            year={2025}
+                            }
+                        ```
+                        ```
+                        @article{zhang2024mowe,
+                            title={MoWE-Audio: Multitask AudioLLMs with Mixture of Weak Encoders},
+                            author={Zhang, Wenyu and Sun, Shuo and Wang, Bin and Zou, Xunlong and Liu, Zhuohan and He, Yingxu and Lin, Geyu and Chen, Nancy F and Aw, Ai Ti},
+                            journal={ICASSP},
+                            year={2025}
+                            }
+                        ```
+                        ```
+                        @article{wang2025advancing,
+                            title={Advancing Singlish Understanding: Bridging the Gap with Datasets and Multimodal Models},
+                            author={Wang, Bin and Zou, Xunlong and Sun, Shuo and Zhang, Wenyu and He, Yingxu and Liu, Zhuohan and Wei, Chengwei and Chen, Nancy F and Aw, AiTi},
+                            journal={arXiv preprint arXiv:2501.01034},
+                            year={2025}
+                            }
+                        ```
+                        ```
+                        @article{he2024meralion,
+                            title={MERaLiON-AudioLLM: Technical Report},
+                            author={He, Yingxu and Liu, Zhuohan and Sun, Shuo and Wang, Bin and Zhang, Wenyu and Zou, Xunlong and Chen, Nancy F and Aw, Ai Ti},
+                            journal={arXiv preprint arXiv:2412.09818},
+                            year={2024}
+                            }
+                        ```
+
+                        """)
+
+
+
+
+
+
+
+def asr_english():
+    st.title("Task: Automatic Speech Recognition - English")
+    
+    sum = ['Overall']
+    dataset_lists = [
+                    'LibriSpeech-Clean', 
+                    'LibriSpeech-Other', 
+                    'CommonVoice-15-EN', 
+                    'Peoples-Speech', 
+                    'GigaSpeech-1', 
+                    'Earnings-21', 
+                    'Earnings-22', 
+                    'TED-LIUM-3', 
+                    'TED-LIUM-3-LongForm', 
+                    ]
+
+    filters_levelone = sum + dataset_lists
+    
+    left, center, _, middle, right = st.columns([0.4, 0.2, 0.2, 0.2 ,0.2])
+    
+    with left:
+        filter_1 = st.selectbox('Dataset', filters_levelone)
+    
+    if filter_1:
+        if filter_1 in sum:
+            sum_table_mulit_metrix('asr_english', ['wer'])
+        else:
+            dataset_contents(dataset_diaplay_information[filter_1], metrics_info['wer'])
+            draw('su', 'asr_english', filter_1, 'wer', cus_sort=True)
+
+
+
+
+
+def asr_singlish():
+    st.title("Task: Automatic Speech Recognition - Singlish")
+
+    sum = ['Overall']
+    dataset_lists = [
+                    'MNSC-PART1-ASR', 
+                    'MNSC-PART2-ASR',
+                    'MNSC-PART3-ASR',
+                    'MNSC-PART4-ASR',
+                    'MNSC-PART5-ASR',
+                    'MNSC-PART6-ASR',
+                    ]
+
+    filters_levelone = sum + dataset_lists
+    
+    left, center, _, middle, right = st.columns([0.4, 0.2, 0.2, 0.2 ,0.2])
+    
+    with left:
+        filter_1 = st.selectbox('Dataset', filters_levelone)
+    
+    if filter_1:
+        if filter_1 in sum:
+            sum_table_mulit_metrix('asr_singlish', ['wer'])
+        else:
+            dataset_contents(dataset_diaplay_information[filter_1], metrics_info['wer'])
+            draw('su', 'asr_singlish', filter_1, 'wer')
+
+
+
+
+def asr_mandarin():
+    st.title("Task: Automatic Speech Recognition - Mandarin")
+
+    sum = ['Overall']
+    dataset_lists = [
+                    'AISHELL-ASR-ZH', 
+                    ]
+
+    filters_levelone = sum + dataset_lists
+    
+    left, center, _, middle, right = st.columns([0.4, 0.2, 0.2, 0.2 ,0.2])
+    
+    with left:
+        filter_1 = st.selectbox('Dataset', filters_levelone)
+    
+    if filter_1:
+        if filter_1 in sum:
+            sum_table_mulit_metrix('asr_mandarin', ['wer'])
+        else:
+            dataset_contents(dataset_diaplay_information[filter_1], metrics_info['wer'])
+            draw('su', 'asr_mandarin', filter_1, 'wer')
+
+    
+
+
+def speech_translation():
+    st.title("Task: Speech Translation")
+    
+    sum = ['Overall']
+    dataset_lists = [
+                        'CoVoST2-EN-ID', 
+                        'CoVoST2-EN-ZH',
+                        'CoVoST2-EN-TA', 
+                        'CoVoST2-ID-EN', 
+                        'CoVoST2-ZH-EN', 
+                        'CoVoST2-TA-EN']
+
+    filters_levelone = sum + dataset_lists
+    
+    left, center, _, middle, right = st.columns([0.4, 0.2, 0.2, 0.2 ,0.2])
+    
+    with left:
+        filter_1 = st.selectbox('Dataset', filters_levelone)
+    
+    if filter_1:
+        if filter_1 in sum:
+            sum_table_mulit_metrix('st', ['bleu'])
+        else:
+            dataset_contents(dataset_diaplay_information[filter_1], metrics_info['bleu'])
+            draw('su', 'ST', filter_1, 'bleu')
+
+
+
+
+def speech_question_answering_english():
+    st.title("Task: Spoken Question Answering - English")
+    
+    sum = ['Overall']
+
+    dataset_lists = [
+                    'CN-College-Listen-MCQ',
+                    'DREAM-TTS-MCQ',
+                    'SLUE-P2-SQA5', 
+                    'Public-SG-Speech-QA', 
+                    'Spoken-SQuAD',
+                     ]
+
+    filters_levelone = sum + dataset_lists
+    
+    left, center, _, middle, right = st.columns([0.4, 0.2, 0.2, 0.2 ,0.2])
+    
+    with left:
+        filter_1 = st.selectbox('Dataset', filters_levelone)
+
+    if filter_1:
+        if filter_1 in sum:
+            sum_table_mulit_metrix('sqa_english', ['llama3_70b_judge'])
+
+        #elif filter_1 in dataset_lists:
+        #    dataset_contents(sqa_datasets[filter_1], metrics['llama3_70b_judge'])
+        #    draw('su', 'SQA', filter_1, 'llama3_70b_judge')
+        
+        else:
+            dataset_contents(dataset_diaplay_information[filter_1], metrics_info['llama3_70b_judge'])
+            draw('su', 'sqa_english', filter_1, 'llama3_70b_judge')
+
+
+
+
+def speech_question_answering_singlish():
+    st.title("Task: Spoken Question Answering - Singlish")
+    
+    sum = ['Overall']
+
+    dataset_lists = [
+              'MNSC-PART3-SQA', 
+              'MNSC-PART4-SQA',
+              'MNSC-PART5-SQA',
+              'MNSC-PART6-SQA',
+              ]
+
+
+    filters_levelone = sum + dataset_lists
+    
+    left, center, _, middle, right = st.columns([0.4, 0.2, 0.2, 0.2 ,0.2])
+    
+    with left: 
+        filter_1 = st.selectbox('Dataset', filters_levelone)
+
+    if filter_1:
+        if filter_1 in sum:
+            sum_table_mulit_metrix('sqa_singlish', ['llama3_70b_judge'])
+        
+        else:
+            dataset_contents(dataset_diaplay_information[filter_1], metrics_info['llama3_70b_judge'])
+            draw('su', 'sqa_singlish', filter_1, 'llama3_70b_judge')
+
+
+def spoken_dialogue_summarization_singlish():
+    st.title("Task: Spoken Dialogue Summarization - Singlish")
+    
+    sum = ['Overall']
+
+    dataset_lists = [
+              'MNSC-PART3-SDS', 
+              'MNSC-PART4-SDS',
+              'MNSC-PART5-SDS',
+              'MNSC-PART6-SDS',
+              ]
+
+
+    filters_levelone = sum + dataset_lists
+    
+    left, center, _, middle, right = st.columns([0.4, 0.2, 0.2, 0.2 ,0.2])
+    
+    with left: 
+        filter_1 = st.selectbox('Dataset', filters_levelone)
+
+    if filter_1:
+        if filter_1 in sum:
+            sum_table_mulit_metrix('sds_singlish', ['llama3_70b_judge'])
+        
+        else:
+            dataset_contents(dataset_diaplay_information[filter_1], metrics_info['llama3_70b_judge'])
+            draw('su', 'sds_singlish', filter_1, 'llama3_70b_judge')
+
+
+
+
+def speech_instruction():
+    st.title("Task: Speech Instruction")
+    
+    sum = ['Overall']
+
+    dataset_lists = ['OpenHermes-Audio', 
+                     'ALPACA-Audio',
+                     ]
+    
+    filters_levelone = sum + dataset_lists
+    
+    left, center, _, middle, right = st.columns([0.4, 0.2, 0.2, 0.2 ,0.2])
+    
+    with left:
+        filter_1 = st.selectbox('Dataset', filters_levelone)
+
+    if filter_1:
+        if filter_1 in sum:
+            sum_table_mulit_metrix('speech_instruction', ['llama3_70b_judge'])
+        else:
+            dataset_contents(dataset_diaplay_information[filter_1], metrics_info['llama3_70b_judge'])
+            draw('su', 'speech_instruction', filter_1, 'llama3_70b_judge')
+
+
+
+
+def audio_captioning():
+    st.title("Task: Audio Captioning")
+
+    filters_levelone = ['WavCaps', 
+                        'AudioCaps',
+                        ]
+    filters_leveltwo = ['Llama3-70b-judge', 'Meteor']
+    
+    left, center, _, middle, right = st.columns([0.4, 0.2, 0.2, 0.2 ,0.2])
+    
+    with left:
+        filter_1 = st.selectbox('Dataset', filters_levelone)
+    with middle:
+        metric = st.selectbox('Metric', filters_leveltwo)
+
+    if filter_1 or metric:
+        dataset_contents(dataset_diaplay_information[filter_1], metrics_info[metric.lower().replace('-', '_')])
+        draw('asu', 'audio_captioning', filter_1, metric.lower().replace('-', '_'))
+
+
+
+
+def audio_scene_question_answering():
+    st.title("Task: Audio Scene Question Answering")
+
+    sum = ['Overall']
+
+    dataset_lists = ['Clotho-AQA', 
+                    'WavCaps-QA', 
+                    'AudioCaps-QA']
+    
+    filters_levelone = sum + dataset_lists
+    
+    left, center, _, middle, right = st.columns([0.4, 0.2, 0.2, 0.2 ,0.2])
+    
+    with left:
+        filter_1 = st.selectbox('Dataset', filters_levelone)
+    
+    if filter_1:
+        if filter_1 in sum:
+            sum_table_mulit_metrix('audio_scene_question_answering', ['llama3_70b_judge'])
+        else:
+            dataset_contents(dataset_diaplay_information[filter_1], metrics_info['llama3_70b_judge'])
+            draw('asu', 'audio_scene_question_answering', filter_1, 'llama3_70b_judge')
+
+
+
+
+def emotion_recognition():
+    st.title("Task: Emotion Recognition")
+
+    sum = ['Overall']
+
+    dataset_lists = [
+                    'IEMOCAP-Emotion', 
+                    'MELD-Sentiment', 
+                    'MELD-Emotion',
+                    ]
+
+    filters_levelone = sum + dataset_lists
+    
+    left, center, _, middle, right = st.columns([0.4, 0.2, 0.2, 0.2 ,0.2])
+    
+    with left:
+        filter_1 = st.selectbox('Dataset', filters_levelone)
+
+    if filter_1:
+        if filter_1 in sum:
+            sum_table_mulit_metrix('emotion_recognition', ['llama3_70b_judge'])
+        else:
+            dataset_contents(dataset_diaplay_information[filter_1], metrics_info['llama3_70b_judge'])
+            draw('vu', 'emotion_recognition', filter_1, 'llama3_70b_judge')
+
+
+
+
+def accent_recognition():
+    st.title("Task: Accent Recognition")
+
+    sum = ['Overall']
+    dataset_lists = [
+        'VoxCeleb-Accent',
+        'MNSC-AR-Sentence',
+        'MNSC-AR-Dialogue',
+        ]
+
+
+    filters_levelone = sum + dataset_lists
+    
+    left, center, _, middle, right = st.columns([0.4, 0.2, 0.2, 0.2 ,0.2])
+    
+    with left:
+        filter_1 = st.selectbox('Dataset', filters_levelone)
+
+
+    if filter_1:
+        if filter_1 in sum:
+            sum_table_mulit_metrix('accent_recognition', ['llama3_70b_judge'])
+        else:
+            dataset_contents(dataset_diaplay_information[filter_1], metrics_info['llama3_70b_judge'])
+            draw('vu', 'accent_recognition', filter_1, 'llama3_70b_judge')
+
+
+
+
+def gender_recognition():
+    st.title("Task: Gender Recognition")
+    
+    sum = ['Overall']
+
+    dataset_lists =  [
+                        'VoxCeleb-Gender', 
+                        'IEMOCAP-Gender'
+                        ]
+
+    filters_levelone = sum + dataset_lists
+    
+    left, center, _, middle, right = st.columns([0.4, 0.2, 0.2, 0.2 ,0.2])
+    
+    with left:
+        filter_1 = st.selectbox('Dataset', filters_levelone)
+    
+    if filter_1:
+        if filter_1 in sum:
+            sum_table_mulit_metrix('gender_recognition', ['llama3_70b_judge'])
+        else:
+            dataset_contents(dataset_diaplay_information[filter_1], metrics_info['llama3_70b_judge'])
+            draw('vu', 'gender_recognition', filter_1, 'llama3_70b_judge')
+
+
+
+
+def music_understanding():
+    st.title("Task: Music Understanding - MCQ Questions")
+    
+    sum = ['Overall']
+
+    dataset_lists =  ['MuChoMusic',
+                      ]
+
+    filters_levelone = sum + dataset_lists
+    
+    left, center, _, middle, right = st.columns([0.4, 0.2, 0.2, 0.2 ,0.2])
+    
+    with left:
+        filter_1 = st.selectbox('Dataset', filters_levelone)
+    
+    if filter_1:
+        if filter_1 in sum:
+            sum_table_mulit_metrix('music_understanding', ['llama3_70b_judge'])
+        else:
+            dataset_contents(dataset_diaplay_information[filter_1], metrics_info['llama3_70b_judge'])
+            draw('vu', 'music_understanding', filter_1, 'llama3_70b_judge')
+
+
+
+
+
+
+
+
+
+
+def under_development():
+    st.title("Task: Under Development")
+    
+
+    dataset_lists =  [
+                      'CNA',
+                      'IDPC',
+                      'Parliament',
+                      'UKUS-News',
+                      'Mediacorp',
+                      'IDPC-Short',
+                      'Parliament-Short',
+                      'UKUS-News-Short',
+                      'Mediacorp-Short',
+
+                      'YouTube ASR: English Singapore Content',
+                      'YouTube ASR: English with Strong Emotion',
+                      'YouTube ASR: Malay English Prompt',
+                      'YouTube ASR: Malay with Malay Prompt',
+                      
+                      'SEAME-Dev-Mandarin',
+                      'SEAME-Dev-Singlish',
+
+                      'YouTube SQA: English with Singapore Content',
+                      'YouTube SDS: English with Singapore Content',
+                      'YouTube PQA: English with Singapore Content',
+
+                      ]
+
+    filters_levelone = dataset_lists
+    
+    left, center, _, middle, right = st.columns([0.4, 0.2, 0.2, 0.2 ,0.2])
+    
+    with left:
+        filter_1 = st.selectbox('Dataset', filters_levelone)
+    
+    dataset_contents(dataset_diaplay_information[filter_1], 'under_development')
+    
+    # = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = =
+
+    '''
+    Show Dataset Examples
+    '''
+
+    # Initialize a session state variable for toggling the chart visibility
+    if "show_dataset_examples" not in st.session_state:
+        st.session_state.show_dataset_examples = False
+
+    # Create a button to toggle visibility
+    if st.button("Show Dataset Examples"):
+        st.session_state.show_dataset_examples = not st.session_state.show_dataset_examples
+
+    if st.session_state.show_dataset_examples:
+        
+        # st.markdown('To be implemented')
+
+        # # if dataset_name in ['Earnings21-Test', 'Earnings22-Test', 'Tedlium3-Test', 'Tedlium3-Long-form-Test']:
+        if filter_1 in []:
+            pass
+        else:
+            try:
+                show_dataset_examples(filter_1)
+            except:
+                st.markdown('To be implemented')
+    # = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = =    
+    
+    if filter_1 in [
+                      'CNA',
+                      'IDPC',
+                      'Parliament',
+                      'UKUS-News',
+                      'Mediacorp',
+                      'IDPC-Short',
+                      'Parliament-Short',
+                      'UKUS-News-Short',
+                      'Mediacorp-Short',
+                      
+                      'YouTube ASR: English Singapore Content',
+                      'YouTube ASR: English with Strong Emotion',
+                      'YouTube ASR: Malay English Prompt',
+                      'YouTube ASR: Malay with Malay Prompt',
+                      
+                      'SEAME-Dev-Mandarin',
+                      'SEAME-Dev-Singlish',
+                      ]:
+        
+        draw('vu', 'under_development_wer', filter_1, 'wer')
+
+    elif filter_1 in [
+        'YouTube SQA: English with Singapore Content',
+        'YouTube SDS: English with Singapore Content',
+        'YouTube PQA: English with Singapore Content',
+        ]:
+        draw('vu', 'under_development_llama3_70b_judge', filter_1, 'llama3_70b_judge')
+
+
+
+
+

show_examples.py

@@ -0,0 +1,193 @@
+import streamlit as st
+import datasets
+import numpy as np
+
+import html
+
+from app.content import displayname2datasetname
+
+def show_dataset_examples(display_name):
+    st.divider()
+    dataset_name = displayname2datasetname[display_name]
+    sample_folder = f"./examples/{dataset_name}"
+    
+    # load dataset
+    dataset = datasets.load_from_disk(sample_folder)
+    
+    for index in range(len(dataset)):
+        with st.container():
+            st.markdown(f'##### Example-{index+1}')
+            col1, col2 = st.columns([0.3, 0.7], vertical_alignment="center")
+
+            # with col1:
+            st.audio(f'{sample_folder}/sample_{index}.wav', format="audio/wav")
+                        
+            if dataset_name in ['CN-College-Listen-MCQ-Test', 'DREAM-TTS-MCQ-Test']:
+                
+                choices = dataset[index]['other_attributes']['choices'] 
+                if isinstance(choices, str):
+                    choices_text = choices
+                elif isinstance(choices, list):
+                    choices_text = ' '.join(i for i in choices)
+                
+                question_text = f"""{dataset[index]['instruction']['text']} {choices_text}"""
+            else:
+                question_text = f"""{dataset[index]['instruction']['text']}"""
+            
+            question_text = html.escape(question_text)
+        
+        with st.container():
+            custom_css = """
+                        <style>
+                        .my-container-table, p.my-container-text {
+                        background-color: #fcf8dc;
+                        padding: 10px;
+                        border-radius: 5px;
+                        font-size: 13px;
+                        # height: 50px;
+                        word-wrap: break-word
+                        }
+                        </style>
+                        """
+            st.markdown(custom_css, unsafe_allow_html=True)
+
+            s = f"""<tr>
+                    <td><b>{html.escape(question_text.replace('(A)', '<br>(A)').replace('(B)', '<br>(B)').replace('(C)', '<br>(C)'))}
+                    </td>
+                    <td><b>{html.escape(dataset[index]['answer']['text'])}
+                    </td>
+            </tr>
+            """
+
+            body_details = f"""<table style="table-layout: fixed; width:100%">
+            <thead>
+                <tr style="text-align: center;">
+                    <th style="width:50%">PROMPT</th>
+                    <th style="width:50%">ANSWER</th>
+                </tr>
+            {s}
+            </thead>
+            </table>"""
+            
+            st.markdown(f"""<div class="my-container-table">
+                            {body_details}
+                            </div>""", unsafe_allow_html=True)
+        
+            st.text("")
+    
+    st.divider()
+    
+
+def show_examples(category_name, dataset_name, model_lists, display_model_names):
+    st.divider()
+    sample_folder = f"./examples/{category_name}/{dataset_name}"
+    
+    dataset = datasets.load_from_disk(sample_folder)
+
+    for index in range(len(dataset)):
+        with st.container():
+            st.markdown(f'##### Example-{index+1}')
+            col1, col2 = st.columns([0.3, 0.7], vertical_alignment="center")
+
+            # with col1:
+            st.audio(f'{sample_folder}/sample_{index}.wav', format="audio/wav")
+                        
+            if dataset_name in ['CN-College-Listen-MCQ-Test', 'DREAM-TTS-MCQ-Test']:
+                
+                choices = dataset[index]['other_attributes']['choices'] 
+                if isinstance(choices, str):
+                    choices_text = choices
+                elif isinstance(choices, list):
+                    choices_text = ' '.join(i for i in choices)
+                
+                question_text = f"""{dataset[index]['instruction']['text']} {choices_text}"""
+            else:
+                question_text = f"""{dataset[index]['instruction']['text']}"""
+
+            question_text = html.escape(question_text)
+            
+            # st.divider()
+            with st.container():
+                custom_css = """
+                            <style>
+                            .my-container-table, p.my-container-text {
+                            background-color: #fcf8dc;
+                            padding: 10px;
+                            border-radius: 5px;
+                            font-size: 13px;
+                            # height: 50px;
+                            word-wrap: break-word
+                            }
+                            </style>
+                            """
+                st.markdown(custom_css, unsafe_allow_html=True)
+
+                model_lists.sort()
+
+                s = f"""<tr>
+                       <td><b>REFERENCE</td>
+                       <td><b>{html.escape(question_text.replace('(A)', '<br>(A)').replace('(B)', '<br>(B)').replace('(C)', '<br>(C)'))}
+                       </td>
+                       <td><b>{html.escape(dataset[index]['answer']['text'])}
+                       </td>
+                </tr>
+                """
+                if dataset_name in ['CN-College-Listen-MCQ-Test', 'DREAM-TTS-MCQ-Test']:
+                    for model in model_lists:
+                        try:
+
+                            model_prediction = dataset[index][model]['model_prediction']
+                            model_prediction = model_prediction.replace('<','').replace('>','').replace('\n','(newline)').replace('*','')
+
+                            s += f"""<tr>
+                                <td>{display_model_names[model]}</td>
+                                <td>
+                                    {dataset[index][model]['text'].replace('Choices:', '<br>Choices:').replace('(A)', '<br>(A)').replace('(B)', '<br>(B)').replace('(C)', '<br>(C)') 
+                                     }
+                                    </td>
+                                <td>{html.escape(model_prediction)}</td>
+                            </tr>"""
+                        except:
+                            print(f"{model} is not in {dataset_name}")
+                            continue
+                else:
+                    for model in model_lists:
+
+                        print(dataset[index][model]['model_prediction'])
+
+                        try:
+
+                            model_prediction = dataset[index][model]['model_prediction']
+                            model_prediction = model_prediction.replace('<','').replace('>','').replace('\n','(newline)').replace('*','')
+
+                            s += f"""<tr>
+                                <td>{display_model_names[model]}</td>
+                                <td>{html.escape(dataset[index][model]['text'])}</td>
+                                <td>{html.escape(model_prediction)}</td>
+                            </tr>"""
+                        except:
+                            print(f"{model} is not in {dataset_name}")
+                            continue
+
+                
+                body_details = f"""<table style="table-layout: fixed; width:100%">
+                <thead>
+                    <tr style="text-align: center;">
+                        <th style="width:20%">MODEL</th>
+                        <th style="width:30%">QUESTION</th>
+                        <th style="width:50%">MODEL PREDICTION</th>
+                    </tr>
+                {s}
+                </thead>
+                </table>"""
+                
+                st.markdown(f"""<div class="my-container-table">
+                                {body_details}
+                                </div>""", unsafe_allow_html=True)
+            
+                st.text("")
+        
+        st.divider()
+
+    
+    

summarization.py

@@ -0,0 +1,127 @@
+import streamlit as st
+import pandas as pd
+import numpy as np
+from streamlit_echarts import st_echarts
+from streamlit.components.v1 import html
+# from PIL import Image 
+from app.show_examples import *
+from app.content import *
+
+import pandas as pd
+from typing import List
+
+from model_information import get_dataframe
+
+info_df = get_dataframe()
+
+
+def sum_table_mulit_metrix(task_name, metrics_lists: List[str]):
+
+    # combine chart data from multiple sources
+    chart_data = pd.DataFrame()
+    for metrics in metrics_lists:
+        folder = f"./results_organized/{metrics}"
+        data_path = f'{folder}/{task_name.lower()}.csv'
+        one_chart_data = pd.read_csv(data_path).round(3)
+        if len(chart_data) == 0:
+            chart_data = one_chart_data
+        else:
+            chart_data = pd.merge(chart_data, one_chart_data, on='Model', how='outer')
+    
+
+    selected_columns = [i for i in chart_data.columns if i != 'Model']
+    chart_data['Average'] = chart_data[selected_columns].mean(axis=1)
+
+    # Update dataset name in table
+    chart_data = chart_data.rename(columns=datasetname2diaplayname)
+    
+    st.markdown("""
+                <style>
+                .stMultiSelect [data-baseweb=select] span {
+                    max-width: 800px;
+                    font-size: 0.9rem;
+                    background-color: #3C6478 !important; /* Background color for selected items */
+                    color: white; /* Change text color */
+                    back
+                }
+                </style>
+                """, unsafe_allow_html=True)
+    
+    # remap model names
+    display_model_names = {key.strip() :val.strip() for key, val in zip(info_df['Original Name'], info_df['Proper Display Name'])}
+    chart_data['model_show'] = chart_data['Model'].map(lambda x: display_model_names.get(x, x))
+
+    models = st.multiselect("Please choose the model", 
+                            sorted(chart_data['model_show'].tolist()), 
+                            default = sorted(chart_data['model_show'].tolist()),
+                            )
+    
+    chart_data = chart_data[chart_data['model_show'].isin(models)].dropna(axis=0)
+
+    if len(chart_data) == 0: return
+
+    # = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = =
+    '''
+    Show Table
+    '''
+    with st.container():
+        st.markdown(f'##### TABLE')
+
+        model_link = {key.strip(): val for key, val in zip(info_df['Proper Display Name'], info_df['Link'])}
+
+        chart_data['model_link'] = chart_data['model_show'].map(model_link) 
+
+        tabel_columns = [i for i in chart_data.columns if i not in ['Model', 'model_show']]
+        column_to_front = 'Average'
+        new_order = [column_to_front] + [col for col in tabel_columns if col != column_to_front]
+        
+        chart_data_table = chart_data[['model_show'] + new_order]
+        
+
+        # Format numeric columns to 2 decimal places
+        chart_data_table[chart_data_table.columns[1]] = chart_data_table[chart_data_table.columns[1]].apply(lambda x: round(float(x), 3) if isinstance(float(x), (int, float)) else float(x))
+
+        if metrics in ['wer']:
+            ascend = True
+        else:
+            ascend= False
+
+        chart_data_table = chart_data_table.sort_values(
+                by=['Average'],
+                ascending=ascend
+            ).reset_index(drop=True)
+        
+        # Highlight the best performing model
+        def highlight_first_element(x):
+            # Create a DataFrame with the same shape as the input
+            df_style = pd.DataFrame('', index=x.index, columns=x.columns)
+            # Apply background color to the first element in row 0 (df[0][0])
+            # df_style.iloc[0, 1] = 'background-color: #b0c1d7; color: white'
+            df_style.iloc[0, 1] = 'background-color: #b0c1d7'
+
+            return df_style
+        
+
+        styled_df = chart_data_table.style.format(
+            {
+                chart_data_table.columns[i]: "{:.3f}" for i in range(1, len(chart_data_table.columns) - 1)
+             }
+        ).apply(
+            highlight_first_element, axis=None
+        )
+
+        st.dataframe(
+                styled_df,
+                column_config={
+                    'model_show': 'Model',
+                    chart_data_table.columns[1]: {'alignment': 'left'},
+                    "model_link": st.column_config.LinkColumn(
+                        "Model Link",
+                    ),
+                },
+                hide_index=True,
+                use_container_width=True
+            )
+            
+    # Only report the last metrics
+    st.markdown(f'###### Metric: {metrics_info[metrics]}')