Datasets Jobs

What is a dataset?
How will my voice be used if I book a job in the datasets category?
Will my voice recording for this type of project be used to synthesize and create new work by the client? Will my voice be used to create works by third parties?
Will my voice recording be used to create, improve, or expand existing speech services models?
Will my voice be synthesized and/or combined with other services or will my voice be cloned?
How can I be sure my voice won't be cloned? What's protecting me?
Are the job's terms subject to change?
What is a brand voice? How is that different from datasets?
If there is a video component to a project, how will my video file be used?
 

What is a dataset?
 

A dataset is a collection of specific information. Voice data collection, also sometimes referred to as speech data collection, involves gathering, measuring, and labeling audio data.

How will my voice be used if I book a job in the datasets category?

The recordings you submit for a datasets job will be used by the client to create or improve models that train machines to better understand and process audio input. Models like this allow companies to improve how well their AI products can understand the human voice, and how conversational those products are, making them more user-friendly. Examples are products like Alexa, Siri, or your in-car navigation. Different languages, accents, styles, and emotions are some of the types of data that are in demand. That more data that can be gathered, the better the AI products become at understanding a broad range of speech.

Will my voice recording for this type of project be used to synthesize and create new work by the client? Will my voice be used to create works by third parties?

No. No new content will be created with your voice by Voices, or the client.

Will my voice recording by used to create, improve, or expand existing speech services models?

Yes. Your voice data from these types of projects will be used to improve existing training models.

Will my voice be synthesized and/or combined with other services or will my voice be cloned?

No. Dataset recordings will NOT be used to create a clone, or replication, of your voice.

Your voice recording will not be heard outside of Voices and/or by any Voices clients. The only use for the files is to improve or create training models for AI products for the client, with no synthesis or replication of your voice.

How can I be sure my voice won't be cloned? What's protecting me?

There are detailed contracts between Voices and the client explicitly outlining what the dataset files can and cannot be used for. This protects both you and Voices and ensures the files are used only under the terms the client has agreed to and paid for. You will also be given a consent form that specifies the scope of the project.

Are the job's terms subject to change?

Not without further contracting and consent between Voices, the client, and you. You have the opportunity to review each job's requirements and licensing and decline anything you're not comfortable with, should that situation arise.
 

What is a brand voice? How is that different from datasets?

A brand voice is a unique voice chosen by a company to represent them across multiple mediums. Examples include the voice assistant in a Mercedes, the voice of Siri, or the voice of brands like Dove and Old Spice.

To become a brand voice a talent needs to be chosen and hired by the client and this is a big deal, as that talent will either be the sole voice for that brand or one of a very select number of voices for that brand. If you are hired by a client to be their brand voice, there will be specific contractual obligations that often include non-compete clauses and exclude you from doing other jobs. This also means you'll be paid much more.

By contrast, datasets are used to create or improve models that train machines to better understand and process audio input. Models like this allow companies to improve how well their AI products can understand the human voice, and how conversational these products are. This is a very different type of job, and talent are typically paid a more modest fee based on the work involved.

If there is a video component to a project, how will my video file be used?

Typically, a video component in a dataset project will serve the purpose of researching mouth movements during specific phrases for future animation projects. The training model based on this information will not create any version resembling you; rather, the model aims to learn general movement and facial expressions. The computer studies the videos to understand how facial expressions convey emotions, and it uses this knowledge to replicate expressions on a screen/avatar. The generated faces will not resemble the people featured in the videos.

 

 

 

 

How would you rate this article?