Soniox (STT)

Real-time speech-to-text via the Soniox WebSocket API. Used by the Subtitle and Live Translation pages when you pick Soniox as the STT method.

Why Soniox

• Real-time — tokens arrive while the speaker is still talking.
• Speaker diarization — per-token speaker labels (e.g. Speaker 1: Hi…).
• In-stream translation — Soniox can translate while transcribing, saving an extra LLM round trip.
• Multi-language — auto-detects the source language even mid-stream.

Get an API key

1. Sign up at https://console.soniox.com
2. Open API keys → Create new API key
3. Copy it (looks like Bearer ...; copy just the token without the Bearer prefix).

Pricing is metered per minute of audio (~$0.005 / minute at the time of writing) — see https://soniox.com/pricing.

Configure in the app

In Settings → Service:

1. Paste the key into Soniox API key → Save

In Settings → Live (for live translation) or Settings → Subtitle (for subtitle generation):

1. Set STT method to Soniox

What it powers

Page	Use Soniox when
Subtitle	Multi-speaker recordings (interviews, panels, meetings) where you want speaker labels in the SRT
Live Translation	Real-time meeting captioning, especially with multiple speakers

Glossary terms

The Soniox WebSocket accepts a glossary of terms to bias recognition toward. The app forwards your active glossary entries automatically — brand names / proper nouns / jargon get recognised more reliably.

Caveats

Online only

Soniox is cloud-only; if your audio is sensitive (medical, legal), use Whisper (local) instead.

Reconnection

The WebSocket auto-reconnects on transient failures with exponential backoff. Long sessions stay connected through brief network blips.

Common errors

Error	Likely cause
AUTH_ERROR	Wrong / expired API key. Re-paste in Settings → Service.
QUOTA_ERROR	Plan limit exceeded.
CONNECTION_ERROR	Network blocked / firewall. Try again from a different network.