{
  "fetched_at": "2026-10-19T06:27:26Z",
  "source_url": "https://www.kaggle.com/competitions/1871-ai-tech-challenge-2023/data",
  "pages": [
    {
      "name": "data-description",
      "content": "# About Dataset\n\n## General Information\n\nThe data for this challenge has been sourced from a Kaggle public dataset (https://www.kaggle.com/datasets/mozillaorg/common-voice?resource=download). Common Voice is a corpus of speech data read by users on the Common Voice website (http://voice.mozilla.org/), and based upon text from a number of public domain sources like user submitted blog posts, old books, movies, and other public speech corpora. Its primary purpose is to enable the training and testing of automatic speech recognition (ASR) systems.\n\n## Structure\n\nThere is an audio zip folder that holds all of the audio clips. The path to these files is included as a value in the train data that accompanies the audio.\n\nEach row of a csv file represents a single audio clip, and contains the following information:\nid - unique identifier for that clip\nfilename - relative path of the audio file within the audio zip folder\ntext - supposed transcription of the audio\nup_votes - number of people who said audio matches the text\ndown_votes - number of people who said audio does not match text\n\nThese final two columns indicate how accurate the training transcription text is. Because some callers may have loud background noise or may be otherwise hard to understand, we have included data that has uncertain transcriptions. This may be used to guide how you train on the data.\n\n## Files\n\n*   **train.csv** - the training set\n*   **test.csv** - the test set\n*   **sample_submission.csv** - a sample submission file in the correct format\n",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "Description",
      "content": "## 1871 – AI Tech Challenge\n## Context\n\nThe company receives millions of calls from its customers every month and the calls are serviced by thousands of agents. The call channel is one of the primary channels through which customers inquire, provide feedback about the product and services, and raise any concerns. So, the call channel and the subsequent interactions with customers are key to delivering an exceptional customer experience and gaining insights into the needs of the customers.\n\nThe customer interaction data obtained through these calls provides valuable data to the operations and other groups within the company. The voice data, when transcribed and mined appropriately, can act as a secondary data source, and can enable the firm to identify new opportunities. \n\nThe following are some of the use cases of transcribed data with near-term business benefits:\n- Assessment of the level of customer satisfaction with a call interaction and the reasons for satisfaction or dissatisfaction.\n- Identification of complaints associated with products or services so that complaints can be addressed efficiently and expeditiously.\n- Understanding the context of the call so that the feedback and opportunities can be summarized and shared with the product development teams.\n- Understanding the topic(s) associated with the calls so that customer service agents can be trained to address questions associated with multiple topics with minimal call transfers.\n\n## Goal of the Competition\nWith the proliferation of Transformer-based-models aiding Natural Language Processing and Understanding, the company is interested in exploring opportunities to enhance current voice to text transcription capability to aid several downstream analytical, reporting, and modeling tasks. Taking advantage of Large Language Model, *******the challenge is to build a reliable voice-to-text transcription model. *******\n\n## Challenge Overview\nThe training data for this challenge consists of audio files and their respective transcribed text. The test data only includes the audio files, that you will need to transcribe and submit. While your submission for the leaderboard will be scored on the accuracy of your transcribed text, we encourage you to take this one step further and perform a topic modeling step on the transcribed text to identify the main topic of the audio, saved as a list of strings of \"topics\" in an additional column. This should not be included in your competition submission, but may be reviewed by the host and company.",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "Evaluation",
      "content": "The evaluation metric for this competition is [Levenshtein Mean](https://medium.com/@ethannam/understanding-the-levenshtein-distance-equation-for-beginners-c4285a5604f0#:~:text=The%20Levenshtein%20distance%20is%20a,change%20one%20into%20the%20other.). The Levenshtein distance is a number that tells you how different two strings are. The higher the number, the more different the two strings are. You are seeking to minimize this metric to rise on the leaderboard.\n\n**Submission Format**\n\nSubmission files should contain two columns: `Id` and `Predicted`. These should both be saved as string, and the `Predicted` column should contain the transcribed in all lowercase with no punctuation, except in the event of a contraction (i.e. that's).",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "rules",
      "content": "We know that in using a publicly available dataset, the correctly transcribed texts can be found. We ask that you refrain from merely submitting those or using them in training, and we will review the code of winning participants to ensure the results were reached organically.\n\nAside from this note, be sure to have fun and challenge yourself!!",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    }
  ]
}