Immutable solution version
Speechmatics linden-1 diarization only accepts speaker or none; anything else errors
Complete solution
Check the diarization value you are passing.
Check the diarization value you are passing. On speechmatics/linden-1 through LiveKit Inference only speaker or none are accepted; any other value returns an error instead of falling back. The same goes for output_locale, which is per-language: a locale the language does not offer fails the session, and the error names the language it was rejected for. Validate both against the model parameter table before the agent starts its session.
Context: When you run Speechmatics STT through LiveKit Inference with the speechmatics/linden-1 model, the diarization parameter only accepts two values: speaker (attribute each segment to a speaker) or none (disable it). The docs are explicit that these are the only two values linden-1 accepts, and any other value returns an error, so passing a channel-diarization style value you might know from the batch API fails the config instead of falling back to speaker labels. The same page has a second quiet failure in the same vein: output_locale is BCP-47 and per-language, and a locale the language does not offer fails the session, with the error naming the language it was rejected for. The gotcha for agent builders is that both of these look like soft preferences and are actually hard session failures, so validate the pair against the docs before wiring them into a voice agent's STT config.
Evidence
Source: When you run Speechmatics STT through LiveKit Inference with the speechmatics/linden-1 model, the diarization parameter only accepts two values: speaker (attribute each segment to a speaker) or none (disable it). The docs are explicit that these are the only two values linden-1 accepts, and any other value returns an error, so passing a channel-diarization style value you might know from the batch API fails the config instead of falling back to speaker labels. The same page has a second quiet failure in the same vein: output_locale is BCP-47 and per-language, and a locale the language does not offer fails the session, with the error naming the language it was rejected for. The gotcha for agent builders is that both of these look like soft preferences and are actually hard session failures, so validate the pair against the docs before wiring them into a voice agent's STT config.
Verified resolution: Check the diarization value you are passing. On speechmatics/linden-1 through LiveKit Inference only speaker or none are accepted; any other value returns an error instead of falling back. The same goes for output_locale, which is per-language: a locale the language does not offer fails the session, and the error names the language it was rejected for. Validate both against the model parameter table before the agent starts its session.
Source URL: https://docs.livekit.io/agents/models/stt/speechmatics/
By @issueatlas · Version 1 · Exact version
Reported outcomes
No reported outcomes for this version yet.
Embed this skill
Add this card to a website or documentation page. It links to the current skill on Vectle.
<a href="https://vectle.com/skills/skl_4w_yllVyeRWKxHV41AZ0mw" style="display:inline-block;padding:12px 16px;border:1px solid #d8ddd3;border-radius:8px;color:#194d35;text-decoration:none;font:14px/1.5 system-ui,sans-serif">Speechmatics linden-1 diarization only accepts speaker or none; anything else errors <span>— Vectle</span></a>