User Instructions

Settings

Haitian Creole
Dataset Reviewer


🛠 Help us improve our AI model to speak Haitian Creole!

Help train our model to speak Haitian Creole by reviewing and validating our dataset. You must speak Haitian/Caribbean Creole fluently to participate in this task. The dataset contains audio recordings of Creole native speakers reading a script. The script is in Caribbean/Haitian Creole and was provided to the speakers in advance. Your contributions will significantly help improve the AI's ability to understand and communicate in Haitian Creole. Currently, there is no AI that can speak Haitian Creole fluently. Your help will change that! Thank you for your support!

To ensure we cover ALL Creole speakers (not just Haitian Creole), each reader has been assign an ID that indicates the region where they grew up (i.e., Haiti, Reunion, Guadeloupe, Martinique, France, and US/Canada). These id's will be used to ensure the AI is well rounded. After you register, you will be assigned a series of IDs to complete. If you have a preference for a specific region, please let us know.

Select the TTS a task:

  • TTS: Text-To-Speech dataset (Helps train the AI to speak Haitian Creole).
  • ASR: Automatic Speech Recognition dataset (Helps train the AI to understand Haitian Creole).

📋 INSTRUCTIONS:

  1. Listen Carefully: Play the audio and verify that the reader has read the Creole text accurately.
  2. Review and Correct: Check the text or transcription for errors or typos. If an issue is found, add a note describing the issue in the Notes section.
  3. IMPORTANT - Trim Noise in Audio's Beginning/Ending: Beginning and trailing spaces in the audio have been trimmed. However, you may encounter recordings with environmental noises (i.e. other occupants speaking in the background, dog barking, etc.). If the noise occurs during the users' reading of the text, then the audio should be excluded by specifying a low Audio-Quality of 1-Poor. If the noise is before or after the reading of the text, then the audio may be used if trimmed. When the audio plays, there is a moving verticle line that depicts the position of the audio.
    1. Listen carefully to the beggining of the audio and ensure that there are no noise (activity in the graph) that occurs before the reader has started reading.
    2. Similarly, and more common, ensure that there are no activity in the graph after the reader has completed reading.
    3. If there are activity (audible or non-audible), use the Trim buttons (1/4 or 1/2 seconds) to remove. Note, that the Trim buttons on the left, trims the beggining and the Trim buttons on the right trims the ending of the audio.
    4. If you've trimmed too much, you can revert one trim at a time. Note, audio trimmings are saved automatically.
    5. At any time, you can Reset all of your changes and start over.
  4. Evaluate Audio Quality: Determine the Audio Quality according to the options provided. 1-Poor, 2-Fair, 3-Good (Default), 4-Very Good, 5-Excellent. Poor audios will be discarded.
  5. Evaluate Reading Level: Determine the Reading Level according to the options provided. 1-Poor, 2-Fair, 3-Good (Default), 4-Very Good, 5-Excellent. Poor readers will be discarded.
  6. Save Your Work: Click on Save to apply and securely record your changes.
  7. Confirm Your Work: After confirming that all is well, check the Verified check box to confirm and lock your changes.

Additional Notes:

  • If necessary, add a Note to provide useful information about the record for the data scientists who will review your changes.
  • If you encounter any problems with the audio, refresh the page to reset the audio player.

Want to Help?

We're looking for fluent Haitian Creole speakers to help validate our dataset.

Register as a Reviewer