👉 How to Choose a Speech Model?
Speech Model Selection
We’ve introduced a clear separation of speech models in the portal. You can choose from the following options:
MARS-8 – Latest model with improved quality
MARS-8 Instruction – Supports instruction-based generation
MARS-7 – Legacy model
How to select a model
Go to Project Settings.
Select your language (very important).
Open the Speech Model dropdown.
Select the model you want to use.
All new re-generations after this switch will be on current model.
Speaker Settings
New options in Speaker Settings:
Clean Reference – Removes background noise from the reference audio to improve voice quality
Maintain Source Accent – Preserves the accent from the original source audio in the generated voice.
When should I use ‘Clean Reference'?
When the reference audio has background noise, echoes, or hum.
When the recording was done in a non-studio environment.
When you want a cleaner, more professional-sounding voice.
When should I use ‘Maintain Source Accent’?
When you want the generated voice to sound closer to the original speaker.
When accent consistency is important (e.g., regional or localized content).
When working with translated or dubbed content but keeping the original accent style.
These options help you fine-tune how closely the generated voice matches the original audio and accent.
ℹ️ Note:
Existing outputs remain unchanged; new regenerations use the current model.