Flux TTS Inline Controls: Pause and Pronunciation
By Steven Van ·
Flux TTS adds inline pauses on batch requests and Early Access IPA pronunciation overrides on batch and streaming, with errors for invalid combinations.
Deepgram's Flux TTS (/v2/speak) now accepts inline pause and pronunciation controls in the text you send, alongside the existing speed setting.
- Pause: an escaped marker such as \{pause:1s\} inserts silence of 500 to 3000 ms in 100 ms steps, with up to 8 per request. It is available on batch (REST) requests only. On streaming, a pause marker fails the connection with DATA-0002.
- Pronunciation (Early Access): inline IPA overrides how a word is said, for example \{"word": "dupilumab", "pronounce": "duːˈpɪljuːmæb"\}. It works on both batch and streaming. Results can vary between generations, so Deepgram advises generating each term several times, on each voice used, before relying on it in production.
Combinations that would degrade audio are rejected with an error rather than producing poor output. Pronunciation cannot be combined with pause or with a speed other than 1.0, and speed is capped at 1.15 when a pause is present. Batch requests return CONTROL_COMBINATION_INVALID or PAUSE_SPEED_CAP_EXCEEDED. Streaming returns ConfigureFailure or DATA-0002.
Applied controls are reported in the dg-pronunciations-applied and dg-breaks-applied response headers on batch, and in controls_applied on streaming. The controls are available in English.