Clarify STT model modes

[ComputelessComputer]

Show batch and realtime badges for Soniox, AssemblyAI, ElevenLabs, and Mistral model variants, and keep realtime aliases safe for batch transcription.

---

Note

Medium Risk
Touches the live WebSocket relay and provider model resolution; ElevenLabs streaming behavior changes for proxied binary audio, though batch misuse of realtime IDs is explicitly guarded.

Overview
STT settings now expose more realtime variants and label models with Batch vs Realtime badges for AssemblyAI, ElevenLabs, Mistral, and Soniox (not only Soniox). Provider pickers add u3-rt-pro, scribe_v2_realtime, and voxtral-mini-transcribe-realtime-2602, with clearer display names (e.g. Universal-3 Pro Streaming).

Batch transcription maps realtime model IDs to each provider’s batch default (scribe_v2_realtime → scribe_v2, Mistral realtime → voxtral-mini-2602) so choosing a streaming model does not break file/batch paths.

Live proxy adds an optional client binary message mapper on the WebSocket relay (single and stereo split). For ElevenLabs, inbound binary audio is rewritten to JSON input_audio_chunk with base64 before upstream send, matching ElevenLabs’ text-based streaming API.

^{Reviewed by Cursor Bugbot for commit 8a9764e. Bugbot is set up for automated code reviews on this repo. Configure here.}

[netlify]

✅ Deploy Preview for old-char canceled.

Name	Link
🔨 Latest commit	8a9764e
🔍 Latest deploy log	https://app.netlify.com/projects/old-char/deploys/6a2bddf36170230008e68ae3

[cursor]

Cursor Bugbot has reviewed your changes using default effort and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 6f00dd2. Configure here.}