Does Brand Voice Lite support SSML?
We don't support SSML for Brand Voices Lite. However our models are natural sounding and should capture speaker disfluencies and semantic nuances really well.
How should the reference and consent audio be accurately recorded?
Ensure the consent and reference audio are recorded under the same voice conditions. Eliminate any background noise, and clearly enunciate the script. The consent audio should be a mono WAV file with LINEAR16 encoding and a 24 kHz sampling rate.
How do I get on the allowlist?
To get allowlist access, ask your account team to fill out the an allowlist intake form.
What are the expected timelines for the process?
- Allowlist access review: 1-2 weeks Model training: 15 minutes Deployment: 15 minutes
What languages are supported?
We only support en-US.
What end-point regions are supported?
We only support the default global endpoint.
I'm satisfied with the model. What steps should I take to prepare for production?
This is a pre-GA feature so we don't support production SLAs. However, the account team needs to file a Responsible AI review request to get the approval for production, due to AI Principle requirements. Approval can take 1-2 weeks. The account team follows these steps:
- Fills out deal review for approval.
- If approved, fills out an internal ticket with the approved document.
- If approved, reviews and grants you permission to use the feature in production.
What is the product pricing?
Review Brand Voices Lite pricing on the pricing details page.
Will there be voice differences between two versions of a Brand Voice Lite model?
Some changes between two versions of a Brand Voice Lite model are to be expected as our technology evolves, even though the models are trained using the same audio data. If you encounter this issue, send us some samples so that we can investigate.
Where can I report issues to Google?
If the issue persists, contact your account representative. Include only one issue per form submission.