Changelog

Word timings

September 2026

Added timestamps=true to the text-to-speech websocket. Each audio message can now carry an alignment giving the start time, duration and character range of every spoken word, for highlighting text as it is played. Offsets index the text you submitted, and times are absolute across the whole stream. See Word Timings.

Voice cloning now rejects a sample it cannot align, with a 400 explaining why, instead of creating a voice that cannot return timings.

Pronunciation control

September 2026

Added <phoneme:...> tags, so you can spell a word out in IPA when it is said wrongly: "Please welcome <phoneme:ˈdeɪtə>, our guest." Available in ten languages. See Pronunciation.

v1.0.0

August 2025

Created a changelog page to track API updates and announcements. Marked the release of version 1.0.0 of the Neuphonic API.