Course audio has to sound the same in module ten as in module one. These voices hold consistent across sessions, handle technical vocabulary through pronunciation dictionaries, and cover nine Indic languages alongside English.

Podcast production is segment based already, which suits generation in short calls. Intros and ad reads generate once and cache, and pairing contrasting voices produces a two-hander without booking two people.

Generation runs at 3.3 times real time, so audio renders faster than it plays and the buffer stays ahead. First audio arrives in about 200 milliseconds over SSE or WebSocket.

Eighty six voices are tagged mature, and twenty also carry the narrative tag. That overlap is the usual starting point for audiobooks and documentary work, where consistency shows over hours rather than seconds.

There is no tone field in the catalog, so these were chosen by listening rather than filtered. Warmth is a property of the voice, not a setting, which makes casting the decision that counts.

Worth saying plainly: nothing in the catalog records pitch, so these six were chosen by ear rather than filtered. There is no pitch control either, so depth is a casting decision.

Guided audio lives on pacing rather than voice character. Speed goes down to 0.5, pauses come from how you write the script, and nine Indic languages are available for regional wellness content.

Experienced screen reader users often run well above normal speed. Speed adjusts from 0.5 to 2.0, and nine Indic languages are covered, where assistive audio is thin across the whole industry.

Explainer scripts get rewritten after the first cut. Generating line by line means a late change costs one line rather than a re-record, and audio arrives in about a second at 44.1 kHz.

Pre-rendering every branch means shipping every branch. WebSocket streaming lets dialogue generate during play instead, so a conversation tree costs what players actually hear rather than what they might.