Tools
Soniox Stt Provider
Soniox async speech-to-text (STT) provider plugin for OpenClaw — audio transcription via media-understanding (stt-async-v5, 60+ languages, speaker diarization)
Install
openclaw plugins install clawhub:guionai/soniox-stt-provider
README
# @guionai/soniox-stt-provider
Soniox async speech-to-text (STT) provider for the OpenClaw media-understanding
audio pipeline. Optional vendor plugin — distributed via ClawHub, not bundled
with core (same pattern as the volcengine provider).
## What it provides
- Async transcription workflow: upload → create job → poll → transcript
- Default model `stt-async-v5`
- Per-model `language` hints (60+ languages; code-switching supported by Soniox)
- Optional speaker diarization (`enableSpeakerDiarization`, default off)
- ~$0.10/hour at Soniox published rates
## Install
```bash
openclaw plugins install clawhub:guionai/soniox-stt-provider
# or from a local checkout (dev):
openclaw plugins install --link ./extensions/soniox
```
Requires OpenClaw >= 2026.8.1.
## Configure
Set the API key in the gateway environment:
```bash
export SONIOX_API_KEY="your_key" # or tools.media.audio.models apiKey
```
Enable as an audio model (example):
```json5
tools: {
media: {
audio: {
models: [
{
provider: "soniox",
model: "stt-async-v5",
capabilities: ["audio"],
language: "zh",
enableSpeakerDiarization: false, // optional; default off
},
],
},
},
}
```
## Permissions & secrets
- Only outbound HTTPS calls to `api.soniox.com` (upload, job, transcript endpoints)
- Single secret: `SONIOX_API_KEY` (falls back to `tts.providers...apiKey` config)
- No inbound listeners, no filesystem writes
## Update
```bash
openclaw plugins update soniox
```
## Smoke test
```bash
openclaw agent -m '转写这段语音并回我文字' # send a voice note, expect text back
# or unit-level:
cd extensions/soniox && npx vitest run # 5 tests: happy path, overrides, diarization, errors
```
tools
Comments
Sign in to leave a comment