Short answer: there’s no server-side option to chunk on a custom delimiter like "---". The docs only expose length-based chunking and explicit flush controls.
What you can do
WebSockets (input streaming):
Control when audio generation starts via generation_config.chunk_length_schedule (character thresholds). (elevenlabs.io)
Force a chunk boundary at your delimiter by sending the text up to '---' and then a message with "flush": true. This clears the buffer and generates immediately. (elevenlabs.io)
Optionally set auto_mode=true to bypass chunk scheduling/buffers when you’re sending complete phrases per delimiter. (elevenlabs.io)
HTTP TTS (non-WebSocket):
Split your input on '---' client-side and make one request per segment.
Use Request Stitching by passing previous_request_ids to keep prosody consistent across your delimiter-based chunks, then concatenate the audio. (elevenlabs.io)
If you actually want a pause at '---' (not a generation chunk), enable SSML parsing and insert <break time="Xs" /> at that point; this affects prosody but does not create streaming chunks. (elevenlabs.io)
References
Generate audio in real-time (WebSockets): chunk_length_schedule, flush. (elevenlabs.io)
WebSocket API reference: auto_mode, enable_ssml_parsing. (elevenlabs.io)
austere oak
glossy fractal