{
  "fetched_at": "2026-10-18T14:15:59Z",
  "source_url": "https://www.kaggle.com/competitions/dlsprint/data",
  "pages": [
    {
      "name": "data-description",
      "content": "The Dataset paper is available at: [https://arxiv.org/abs/2206.14053](https://arxiv.org/abs/2206.14053). \n\n\nThe dataset for this competition comprises about 400 hours of Bengali audio clips collected through Common Voice. Your task is to generate the Bengali transcripts for the audios. \n\nWhile the training data includes some metadata about the speakers, such metadata will not be provided for any test data. The full dataset (including the Phase 1 Part 1 test set) is publicly available on the Common Voice website. **The participants are prohibited from using the test dataset for training and any such attempt will lead to disqualification.**\n \n## Files\n\n*   **train.csv** - the training set\n*   **validation.csv** - the validation set\n*   **sample_submission.csv** - a sample submission file in the correct format\n* **train_files/** - the training audio files\n* **validation_files/** - the validation audio files\n* **test_files/** - the test audio files. You need to predict the sentence from these audio files.\n\n\n## Columns\n\n*   `client_id` - unique identifier for each speaker\n*   `path` - name of the audio file in corresponding folder\n*   `sentence` - the transcript of the audio file\n*   `up_votes` - reviewer up votes\n*   `down_votes` - reviewer down votes\n*   `age` - age of the speaker\n*   `gender` - gender of the speaker\n*   `accents` - accents of the speaker\n*   `locale` - locality of the speaker\n\n\n\n",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "Description",
      "content": "Bengali is the fifth most spoken of all native languages all over the world. But so far very little work has been done on Bengali Speech Transcription. Considering the large audience and far-reaching opportunities, there’s significant business and educational interest in developing AI that can be used in **𝐁𝐞𝐧𝐠𝐚𝐥𝐢 𝐀𝐒𝐑** (Automatic Speech Recognition). We, at the 𝐃𝐞𝐩𝐚𝐫𝐭𝐦𝐞𝐧𝐭 𝐨𝐟 𝐂𝐨𝐦𝐩𝐮𝐭𝐞𝐫 𝐒𝐜𝐢𝐞𝐧𝐜𝐞 𝐚𝐧𝐝 𝐄𝐧𝐠𝐢𝐧𝐞𝐞𝐫𝐢𝐧𝐠, 𝐁𝐔𝐄𝐓, in partnership with 𝐁𝐞𝐧𝐠𝐚𝐥𝐢.𝐀𝐈, are glad to present the very first Bengali ASR competition of its kind, 𝐃𝐋 𝐒𝐩𝐫𝐢𝐧𝐭, as a part of BUET CSE Fest 2022.\n\nFor training the model, we will be using a crowdsourced sentence-level automatic speech recognition corpus collected on the Mozilla Common Voice platform. The details of the dataset can be found here: [https://arxiv.org/abs/2206.14053](https://arxiv.org/abs/2206.14053). \n\n**This competition is open for all, not only for undergraduates. Anyone from any institution can participate.**\n\n##Competition Website \n\nImportant information and FAQ on the competition can be found here: [https://buetcsefest2022.com/dlsprint/](https://buetcsefest2022.com/dlsprint/). \nYou can even join this facebook group for relevant updates: [https://www.facebook.com/groups/1185590455630397](https://www.facebook.com/groups/1185590455630397).\nRegistration link: [https://tiny.cc/dlsprint22](https://tiny.cc/dlsprint22). **Registration is mandatory to participate in Phase 1 Part 2 and Phase 2 rounds.**\n\n##Competition Schedule \n\n**Phase 1 Part 1:** Will be held in Kaggle from July 1st to August 25th (12:00 AM GMT+6). Participants are expected to build and train their model during this phase.\n**Phase 1 Part 2:** Will be held in Kaggle on August 25th. **Only registered teams will be allowed.** Participants will infer their model on a new dataset. The top 15 teams from this round will be selected for Phase 2.\n**Phase 2:** Will be held onsite at the Department of CSE, BUET, on September 2nd. Only invited participants can participate. Participants will infer their model on a new dataset and present their methods live in front of the judges’ panel. \n\n##Important\n**All teams will submit working Kaggle notebooks for Train and Test with appropriate instructions and documentations for reproducibility after Phase 1 Part 1.**\n\n##Acknowledgements\n\nEBLICT Project, BCC, ICT Division, Bangladesh and IEEE Computer Society BUET Student Branch Chapter are jointly hosting the event with the organizers. The event is also supported by [Incepta Solutions Limited](https://www2.inceptasolutions.com/).",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "Evaluation",
      "content": "Predictions in phase1 are evaluated by [Levenshtein Mean Distance](https://en.wikipedia.org/wiki/Levenshtein_distance) between the predicted sentence and the ground truth sentence. The Levenshtein distance between two sentences is the minimum number of single-character edits (insertions, deletions, or substitutions) required to change one sentence into the other\n\nThe Levenshtein Distance between two strings *a, b* having length |a| and |b| correspondingly is given by *lev(a, b)*\n![Levenshtein Distance](https://wikimedia.org/api/rest_v1/media/math/render/svg/6224efffbe9a4e01afbddeeb900bfd1b3350b335)\n\nNow, Levenshetin Mean for **n** samples, where **gt[i]** and **pred[i]** indicates ground truth and prediction of sample **i**, is calculated by,\n\n$$levenshteinMean=\\frac{\\sum_{i=1}^{n} lev\\(gt\\[i\\], pred\\[i\\] \\)}{n}$$\n\n**Note:** For this competition, we will consider finding Levenshtein Mean among the Unicode token values of ground truth and predicted sentence. \n\n**All teams will submit working Kaggle notebooks for Train and Test with appropriate instructions and documentations for reproducibility after Phase 1 Part 1.**\n\n## Submission Format\n\n\nThe submission files should contain two columns: `path` and `sentence`.  You will need to predict the sentence for each file in the `test_files` folder. \n\n\nThe submission file should contain a header and have the following format:\n```\npath, sentence\ncommon_voice_bn_31675220.mp3, এছাড়াও নিউজিল্যান্ড এ ক্রিকেট দলের হয়েও খেলছেন তিনি।\ncommon_voice_bn_31513116.mp3, এছাড়াও নিউজিল্যান্ড এ ক্রিকেট দলের হয়েও খেলছেন তিনি।\n```",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "rules",
      "content": "Registration link: [https://tiny.cc/dlsprint22](https://tiny.cc/dlsprint22). Registration is mandatory to participate in **Phase 1 Part 2** and **Phase 2** rounds.\n\n- **Only one account per participant in Kaggle will be allowed.**\n- **Internet access should be disabled for any solution.**\n- **All teams will submit working Kaggle notebooks for Train and Test with appropriate instructions and documentation for reproducibility after Phase 1 Part 1**. \n- Use of publicly available pretrained models are allowed and the participants must disclose the data used to pretrain the models. Furthermore, participants must directly use these models in their training, they cannot additionally pretrain their models in another dataset.\n\n###Team guidelines:\nTeams must follow this guidelines in order to qualify for **Phase 1 Part 2** and **Phase 2**.\n\n  - The maximum team size is 4.\n  - In a team, there can be a maximum of 2 people who have completed their undergrad.\n  - For overseas participants, a minimum of 1 person should be present in Bangladesh during the Phase 2 round.\n  - A minimum of 75% of team members must belong to the 18/19/20 batch from BUET to qualify as a BUET rising team.\n  - Teams can merge only in Phase 1 Part 1. In the event of a team merger, team registration should be updated. The registration update link will be shared later.\n  - No private code sharing outside teams.\n\n### Evaluation:\n\n**Phase 1 Part 1 score is for benchmarking and validating participants' models and will also count towards the final evaluation score.**  The **Phase 1 Part 2** and **Phase 2** test datasets may contain samples with a slight domain shift (vocabulary) to encourage generalized models. A full evaluation will be based on the following factors (these are subject to change based on the organizers’ decision. Also, the weights of these factors will not be revealed to the participants):\n\n- Phase 1 score\n- Phase 2 score\n- Reproducibility\n- Inference time\n- Model size\n- Explainability\n- Presentation\n\n### Competition Data:\n\n- **DATA ACCESS AND USE**: Competition Use, Academic Use, and Non-Commercial Use.\n- **NO EXTERNAL DATASET IS ALLOWED**. Manual labeling of the test dataset is prohibited and will lead to disqualification.\n\n### General Rules\n\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- **Winner license**: The winners will grant the competition hosts and sponsors the following license:\nYou 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.\n- **The DL Sprint Organizing Committee reserves the right to cancel or postpone the event for any reason.**\n- **The organizers serve the right to disqualify any team.**\n- **Winners will be selected at the sole discretion of the DL Sprint Organizing Committee. All participants must conform to the terms and conditions. The DL Sprint Organizing Committee’s decisions in this regard will be final, and no queries or appeals thereof will be entertained.**\n\n",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "Timeline",
      "content": "- **August 25, 2022** - Phase 1 Final submission deadline.\n- **August 26, 2022** - Phase 1 Code submission deadline.\n- **September 2, 2022** - Phase 2 evaluation. (only invited teams can participate)\n\nAll deadlines are at 11:59 PM GMT+6 on the corresponding day unless otherwise noted. The competition organizers reserve the right to update the contest timeline if they deem it necessary.\n\n**For details visit [our competition website](https://buetcsefest2022.com/dlsprint/).**",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    }
  ]
}