Data contribution validation

Validators have the privilege of accessing the raw data sources, which is fundamental for assessing the data's quality and relevance.

On the proposal page, Validators will find a link to the dataset they need to review. This link is key to accessing the data for validation purposes. To access the data, Validators must connect their wallet.

By following the link, dataset will be downloaded and Validators can validate them on their devices.

By accepting this dataset, validators agree that it adheres to the following guidelines and is ready for use in fine-tuning a model.

Checklist: Validate a Voice Data

Below is the checklist to validate a Voice Data. It should mark all the checklist to pass the validation, if relevant.

• Voice is not corrupted.

• Voice is a complete recording.

• Voice is ethically sourced and does not infringe on copyrights.

• Voice are clear without significant background noise or distortions. There should not be able camera flashing sound, clicking sounds, white noise or noise in the background that could affect the voice training.

• Voice volume levels are consistent.

• Language used in Voice is accurate and consistent.

• Voice belongs to only one character, there should not be other voices as this will affect the voice training.

• Voice should sound like the character, or better (if Voice Core Maturity Score is low)

• There is no similar voice data in the database. More of the similar one will not improve the model.

• There is new phrase/words/tone in this voice data that worth incorporating as it will improve the Voice Core.

• (Applicable only when this piece of Voice Data is contributed to Domain Expertise Core) The facts are accurate.

Checklist: Validate a Text-based Data

Below is the checklist to validate a Text-based Data. It should mark all the checklist to pass the validation, if relevant.

• Data is not corrupted.

• Data is ethically sourced and does not infringe on copyrights.

• Data is readable and understandable.

• Phrases and sentences do not have grammar mistake or spelling error.

• Formatting and style used is accurate and consistent.

• For labelled dataset, the labels are accurate.

• Language used in the dataset is consistent.

• Texts do not include any unecessary biases (gender, racial, societal) unless it is required.

• (Applicable only when this piece of Text Data is contributed to Domain Expertise Core) The facts are accurate.

• There is no similar data in the database. More of the similar one will not improve the model.

• There is new phrase/words that worth incorporating as it will improve the model.