{
  "fetched_at": "2026-10-18T14:12:32Z",
  "source_url": "https://www.kaggle.com/competitions/ben10/data",
  "pages": [
    {
      "name": "data-description",
      "content": "The recordings are 16kHz *.wav audio files. The data are coming from 373 different speakers from 10 regions in Bangladesh (rangpur, kishoreganj, narail, chittagong, narsingdi, tangail, habiganj, barishal, sylhet, sandwip) There are in total 13610 samples in the train set (63:08:42 hrs) and 1703 samples in the phase 1 test set (8:01:11 hrs).\n\n<> notation is present throughout the train and test sets where the there exists a word, but not intelligible (due to noise, slurred speech, overlap etc). Newline operators (\\n) are present only in the training data but not in the test ground truth files. \n\nThe unicode of the validation (Phase 1 test set)/test (Phase 2 test set) dataset ground truth is normalized using [bnunicodenormalizer](https://github.com/mnansary/bnUnicodeNormalizer). \n\n## Files\n\n* The **16_kHz_train_audio** folder contains the train files, and the **16_kHz_valid_audio** folder contains the phase 1 test set files.  \n*   **train.csv** - the training set transcriptions\n*   **sample_submission.csv** - a sample submission file in the correct format.\n\n## Columns\n\n*   `id` - wav file name. (lorem1.wav)\n*   `sentence` - transcript of file.",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "Description",
      "content": "## Challenge Overview\nThis project of Bengali.AI bridges the gap between the modern speech recognition system and resource development. The primary aim of this competition is to come up with multiple solution systems that can enable the computer to better understand different ways people speak in various regions, focusing on Bengali, a language not extensively supported in terms of technology for its local dialects.\n\n## Challenge Highlights\nOur findings revealed that current advanced technology struggles to accurately understand these local dialects, both when using it directly (zero-shot) and when trying to adapt or fine-tune it for better performance. This challenge exists despite significant efforts and advancements in speech recognition technologies, highlighting a crucial area for further research and development.\n\n## Why Join the Challenge?\nTo contribute to this field and aid in the development of more inclusive and effective speech recognition systems, we've made our dataset and the tools developed during this study freely available. We hope that these resources will serve as a foundation for future work, aiming to improve how computers understand spoken language in all its variety, including the rich tapestry of Bengali dialects. This project not only sheds light on the existing limitations of speech recognition technology in handling dialectal variations but also provides a valuable benchmark resource for testing and enhancing the capabilities of automatic speech recognition systems under resource-constrained conditions.",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "Evaluation",
      "content": "Submissions are evaluated by a mean Word Error Rate, proceeding as follows:\n\n- The WER is computed for each instance in the test set.\n\n- The WERs are averaged within domains (dialects), weighted by the number of words in the sentence.\n\n- The dialect WERs are summed using varying weights.\n\nThe dialect weights are given below:\ndomain_weights = {\n        'Barishal': 0.125,\n        'Chittagong': 0.083,\n        'Habiganj': 0.125,\n        'Kishoreganj': 0.083,\n        'Narail': 0.083, \n        'Narsingdi': 0.083,\n        'Rangpur': 0.083,\n        'Sylhet': 0.125,\n        'Sandwip': 0.125,\n        'Tangail': 0.083,\n    }\n\nThe full kaggle evaluation code is published [**here**](https://www.kaggle.com/code/reasat/regional-asr-wer)",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "rules",
      "content": "# Rules\n\n## Registration\n- 📝 Registration form link : https://forms.gle/LqrFb8ktZH11RWaf7 \n- 📆 Deadline for registration: **April 24, 2024**\n\n## Team Requirements\n- A team must consist of at most **3 members.**\n- The members of a team can be from different universities. Cross university team is allowed.\n- Open for both undergraduate and graduate students as well as working professionals\n- For overseas participants, **at least 1 person** should be present in Bangladesh during the final presentation.\n\n## Rules for the Competition\n- Registration is mandatory to participate in all the phases of the contest.\n- **Only one account** per participant in Kaggle will be allowed.\n- Participants are allowed to use external data, but the data has to be disclosed publicly before the final round.\n- The use of the internet is allowed for research and reference, but API calls are not permitted.\n- All teams must submit working Kaggle notebooks for training and inference (and the models too), including appropriate instructions and documentation for reproducibility. The inference notebook must run on Kaggle Notebooks.\n- Teams are required to make all scripts, data, and reports publicly available.\n- No private code sharing outside teams.\n\n## Phases\n- The teams will be assessed on the leaderboard score from Phase 1 and top 10 teams from Phase 1 private leaderboard will be considered for Phase 2. \n- The phase 2 evaluation will be an in-person event where the teams will be judged based on their hidden evaluation score, presentation, report, and code reproducibility and novelty. \n- **Scoring rubric**:\nPhase 1 leaderboard (Public): 10%\nPhase 1 leaderboard (Private): 20%\nPhase 2 hidden leaderboard: 40%\nPaper Report (PDF): 10%\nPresentation: 10%\nInsights and findings: 10%\n\n## Evaluation\nThe evaluation metric for the competition is Word Error Rate (WER). The dialect weights are given below:\ndomain_weights = {\n        'Barishal': 0.125,\n        'Chittagong': 0.083,\n        'Habiganj': 0.125,\n        'Kishoreganj': 0.083,\n        'Narail': 0.083, \n        'Narsingdi': 0.083,\n        'Rangpur': 0.083,\n        'Sylhet': 0.125,\n        'Sandwip': 0.125,\n        'Tangail': 0.083,\n    }\n\n\n## General Rules\n- The **IUTCS Datathon Organizing Committee** retains the authority to cancel or reschedule the event under any circumstances. The organizers serve the right to disqualify any team.\n- The selection of winners will be made solely at the discretion of the **IUTCS Datathon Organizing Committee**. All participants are required to adhere to the terms and conditions. Decisions made by the **IUTCS Datathon Organizing Committee** in this matter will be conclusive, and no requests for clarification or appeals will be considered.\n- The participants must only use open source code licensed under an Open Source Initiative-approved license (see www.opensource.org) that in no event limits commercial use of such code or model containing or depending on such code.\n\n### Winner license\n- The winners will grant the competition hosts and sponsors the following license:\n    - You hereby license and will license your winning submission and the source code used to generate the submission under an Open Source Initiative-approved license (see www.opensource.org) that in no event limits commercial use of such code or model containing or depending on such code. For generally commercially available software that you used to generate the submission that is not owned by you but that can be procured by the Competition Sponsor without undue expense, you do not need to grant the license in the preceding sentence for that software. You must also make your solution public.\n- The Competition hosts and sponsors and its affiliates may use your name and likeness for advertising and promotional purposes without additional compensation unless prohibited by law.\n- You warrant that your submission is your own original work and, as such, you are the sole and exclusive owner and rights holder of the submission, and you have the right to make the submission and grant all required licenses. To the maximum extent permitted by law, you indemnify and agree to keep indemnified Competition Entities at all times from and against any liability, claims, demands, losses, damages, costs, and expenses resulting from any of your acts, defaults or omissions and/or a breach of any warranty set forth herein.",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "abstract",
      "content": "**Goal Of The Competition**\n\nThe prime goal of this competition is to come up with a solution system to transcribe Bengali speech with various regional dialects following the orthography set by linguists. The speech corpus for this competition has been developed by Bengali.AI from one of their ongoing large-scale research projects. The corpus consists of spontaneous speech on everyday topics from **373 individuals** with various regional dialects from **ten** different geographical locations such as **Rangpur, Kishoreganj, Narail, Chittagong, Narsingdi, Tangail, Barishal, Habiganj, Sylhet & Sandwip.**  The cumulative length of this entire speech corpus is over **79+ hours**. Your efforts could improve the Bengali speech recognition technology using this unique speech recognition dataset for the very first time in history which is dealing with the regional speech domain. In addition, your submission will be among the first open-source speech recognition methods for Bengali.\n\n**Footnote:** This dataset is the subset version of the dataset from an ongoing research project conducted by Bengali.AI which BRAC University is currently funding.\n\n##Sponsors\n\n**Brainstation 23**\nBrainStation23, a globally leading one of the best software development company in Bangladesh. It helps to transform your business into the digital sphere.\n\n**APSIS**\nIt is a company where professionals from both technical and functional fields group together with an objective of providing fit-for-purpose business solutions. It realizes the importance of functional knowledge and its impact in developing business solutions. We constantly strive to be a leading technology firm with profound business and functional knowledge.\n\nBoth companies, being strong advocates for open-sourced research, have been generous enough to sponsor the different aspects of this competition. \n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F1515169%2F832b3be97ed834213a3a60ce2c421a14%2FUntitled%20Diagram.drawio.png?generation=1713863357086185&alt=media)\n\n##Powered by\n\n**Bengali.AI**\nBengali.AI is a community that works to create  Bangla AI language technologies & open-source datasets for research. \n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F7662209%2F02ea43bdf4b2c647e404d429dfa1153a%2Fbengali.ai_400x400.png?generation=1714348850731458&alt=media)\n\n**BRAC University**\nBRAC University is currently funding this research named \"Cross-domain 10k hour Bengali Speech Corpus for state-of-the-art Bengali Speech Recognition, Diarization and Synthesis\" under the \"Brac University Research Seed Grant Initiative (RSGI)\" research grant.\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F7662209%2Fb44ffbd42f6fdc621a2bb2b51d7b1c3b%2Fbracu_logo_ai%20Converted.png?generation=1714348863039559&alt=media)\n\n\n\n",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    }
  ]
}