Deepgram
Use Deepgram speech-to-text models in Vapi
Deepgram provides speech-to-text for Vapi voice agents.
You can use Deepgram through Vapi’s default integration, or connect your own account in Integrations.
Supported capability
Speech-to-text
Set transcriber.provider to deepgram and transcriber.model to a Deepgram model.
For additional API configuration options, review the DeepgramTranscriber fields in the Create Assistant API reference.
Feedback filter
When the caller repeats what the assistant just said, Vapi drops that part of the transcript so the assistant doesn’t react to its own words echoed back. To keep that speech and forward it to the model instead, set transcriber.feedbackFilterEnabled to false. The filter is on by default.
Supported languages
The matrix shows the language and model combinations currently supported by Vapi.
The Nova 3 column applies to Nova 3 and Nova 3 General, and the Nova 2 column applies to Nova 2 and Nova 2 General. Specialized Nova 2 includes Meeting, Phone Call, Finance, Conversational AI, Voicemail, Video, Medical, Drive-Thru, and Automotive.
Multilingual transcription
For Nova 3, Nova 3 Medical, or Nova 2, set transcriber.language to multi. For Flux, select flux-general-multi and omit transcriber.languages to use automatic language detection. You can also provide a languages array to give Flux language hints for multilingual or code-switching conversations. See Deepgram’s Flux language prompting documentation.