Soniox
Configure real-time, multilingual speech-to-text for Vapi assistants
Soniox is a real-time speech-to-text (STT) provider that transcribes caller audio for Vapi assistants. See Supported models for the models you can use.
What is Soniox?
Soniox provides low-latency, multilingual transcription and semantic endpointing for voice assistants.
Speech-to-text
Set transcriber.provider to soniox and transcriber.model to one of the supported models. Set transcriber.languages to an empty array for automatic language detection.
For additional API configuration options, review the SonioxTranscriber fields in the Create Assistant API reference.
Supported models
Automatic language detection
Set transcriber.languages to an empty array to let Soniox automatically detect and transcribe any supported language, including code-switching within a conversation. You can provide expected language codes, such as ["en", "es"], instead. Vapi defaults transcriber.languageHintsStrict to true, which restricts transcription to the provided languages. Set it to false to use the values as hints while allowing other supported languages. See Soniox language hints for provider details.
Supported languages
The following languages are currently accepted by Vapi for Soniox. See Soniox supported languages for provider details.
Low-confidence filter
You can discard transcripts that Soniox returns with low confidence, so the assistant does not respond to uncertain speech. In the dashboard, open the transcriber Advanced settings and turn on the Low-confidence filter to set a threshold. Via the API, set transcriber.confidenceThreshold to a value between 0 and 1.
Confidence is the mean of the per-token scores for the transcript. Any final transcript below the threshold is discarded. When confidenceThreshold is unset, nothing is discarded by this setting. See the SonioxTranscriber fields for the full configuration.
Echo filtering
With stt-rt-v6, Vapi sends the assistant’s speech to Soniox as context. Soniox uses it to filter the assistant’s own voice out of the caller’s transcript, such as echo from a speakerphone, and to improve caller transcription. This is on by default. To turn it off, set transcriber.agentTextInputEnabled to false. Other Soniox models don’t support this setting.
Related
- Configure a transcriber fallback plan.
- Review the Soniox speech-to-text models.