Model updates and feature changes
- Embeddings support: voice embeddings are retired in our latest models. Dropping them measurably improved generation quality and run-to-run stability.
- Voice controls: Pace and Emotion are paused while we harden the newest model. If you still need them, call the API with
aster-2-2025-03-07and follow the migration note in the reference. - Voice cloning: current models support the
similaritymethod only, which returns the closest match to the source take. - Voice designer: discontinued — it depended entirely on embeddings.
