Feedback

Deepgram

Deepgram is built on end-to-end deep learning, delivering some of the fastest and most accurate real-time transcription available. Its Nova-2 model offers industry-leading word error rates at significantly lower cost than legacy providers. Best suited for English-heavy use cases where speed, accuracy, and cost-efficiency are top priorities.
Usage
To use Deepgram as the STT engine, pass the following JSON in the STTConfig field of the StartAIConversation API:
// json — STTConfig
{
  "Language": "en",
  "VadSilenceTime": 1000,
  "CustomParam": "{\"STTType\":\"deepgram\",\"ApiKey\":\"<your_deepgram_api_key>\",\"Model\":\"nova-3\"}"
}
For the complete STTConfig parameter reference, see the STTConfig configuration guide.
Parameter reference
STTConfig fields
The following fields are part of STTConfig. For the full definition, see STTConfig.
Field
Type
Required
Description
Language
String
No
Language code. See Deepgram language support.
VadSilenceTime
Integer
No
VAD silence duration (ms). See STTConfig.
CustomParam fields
CustomParam is not part of the standard STTConfig fields. It is only required when using a third-party STT engine, and is used to pass the service provider's authentication parameters.
Field
Type
Required
Description
STTType
String
Yes
Fixed value: "deepgram".
ApiKey
String
Yes
Your Deepgram API key. Obtain from Deepgram Console.
Model
String
Yes
STT model to use, e.g. "nova-3". See Deepgram Models & Languages.
For more details on Deepgram, see the Deepgram documentation.
Next step: Configure your LLM provider﻿
﻿

Field	Type	Required	Description
`Language`	String	No	Language code. See Deepgram language support.
`VadSilenceTime`	Integer	No	VAD silence duration (ms). See STTConfig.