Voice agents
Transcriber
Choose how your agent hears callers, set the language and reduce background noise.
Settings
- Providerenumrequired
- Speech-to-text provider. Default
deepgram. - Modelstring
- Provider model. Default
nova-3. - Languagestringrequired
- Language code, for example
en,es,fr. Defaulten. - Endpointingboolean
- Deepgram only. Uses provider endpointing (300 ms) to detect the end of speech.
- Background denoisingboolean
- Filters background noise such as traffic or TV before transcription.
Tips
- Set the language to the language your callers actually speak. A wrong language setting is the most common cause of poor transcripts.
- Turn on background denoising for callers in cars, shops and outdoors.
- Add product names, people's names and jargon to the prompt so the model can recover from small transcription errors.
Last updated November 2026 · Found a problem? Email support@katexs.com
