Configure the Voice Agent — Deepgram

Voice Agent

To configure the voice agent, you’ll need to send a Settings Configuration message immediately after connection. Below is a detailed explanation of the configurations available.

Configuration Parameters

Parameter	Description
`audio.input`	The speech-to-text audio media input you wish to send. The audio must be uncontainerized and have the following encodings: - `Linear16` - `alaw` - `mulaw`
`audio.output`	The text-to-speech audio media output you wish to receive. See options
`agent.listen.model`	The Deepgram speech-to-text model to be used. See options
`agent.listen.keyterms`	Keyterms you want increased recognition for. See more
`agent.think.model`	Defines the LLM Model to be used. See options
`agent.think.provider.type`	Defines the LLM Provider See options
`agent.think.instructions`	Defines the System Prompt for the LLM
`agent.think.functions`	Pass functions to the agent that will be called throughout the conversation if/when needed as described per function. See options
`agent.speak.model`	The Deepgram text-to-speech model to be used. See options
`context.messages`	Used to restore existing conversation if websocket connection break.
`context.replay`	Used to to replay the last message, if it is an assistant message.

Full Example

Below is an in-depth example showing all the fields available fields for SettingConfigurations highlighting default values and optional properties.

JSON

1 {
2   "type": "SettingsConfiguration",
3   "audio": {
4     "input": { // optional. defaults to 16kHz linear16
5       "encoding": "",
6       "sample_rate": 24000 // defaults to 24k
7     },
8     "output": { // optional. see table below for defaults and allowed combinations
9       "encoding": "",
10       "sample_rate": 24000, // defaults to 24k
11       "bitrate": 0,
12       "container": ""
13     }
14   },
15   "agent": {
16     "listen": {
17       "model": "", // optional. default 'nova-3'
18       "keyterms": [] // optional, only available on nova 3 models
19     },
20     "think": {
21       "provider": {
22         "type": "" // see `LLM providers and models` table below
23       },
24       "model": "", // see `LLM providers and models` table below
25       "instructions": "", // optional (this is the LLM System prompt)
26       "functions": [
27         {
28           "name": "", // function name
29           "description": "", // tells the agent what the function does, and how and when to use it
30           "url": "", // the endpoint where your function will be called
31           "headers": [ // optional. Http headers to pass when calling the function. Only supports 'authorization'
32             {
33               "key": "authorization",
34               "value": ""
35             }
36           ],
37           "method": "post", // the http method to use when calling the function
38           "parameters": {
39             "type": "object",
40             "properties": {
41               "item": { // the name of the input property
42                 "type": "string", // the type of the input
43                 "description":"" // the description of the input so the agent understands what it is
44               }
45             },
46             "required": ["item"] // the list of required input properties for this function to be called
47           }
48         }
49       ]
50     },
51     "speak": {
52       "model": "" // default 'aura-asteria-en' for other providers see TTS Models documentation
53     }
54   },
55   "context": {
56     "messages": [], // LLM message history (e.g. to restore existing conversation if websocket connection breaks)
57     "replay": false // whether to replay the last message, if it is an assistant message
58   }
59 }

1	{
2	"type": "SettingsConfiguration",
3	"audio": {
4	"input": { // optional. defaults to 16kHz linear16
5	"encoding": "",
6	"sample_rate": 24000 // defaults to 24k
7	},
8	"output": { // optional. see table below for defaults and allowed combinations
9	"encoding": "",
10	"sample_rate": 24000, // defaults to 24k
11	"bitrate": 0,
12	"container": ""
13	}
14	},
15	"agent": {
16	"listen": {
17	"model": "", // optional. default 'nova-3'
18	"keyterms": [] // optional, only available on nova 3 models
19	},
20	"think": {
21	"provider": {
22	"type": "" // see `LLM providers and models` table below
23	},
24	"model": "", // see `LLM providers and models` table below
25	"instructions": "", // optional (this is the LLM System prompt)
26	"functions": [
27	{
28	"name": "", // function name
29	"description": "", // tells the agent what the function does, and how and when to use it
30	"url": "", // the endpoint where your function will be called
31	"headers": [ // optional. Http headers to pass when calling the function. Only supports 'authorization'
32	{
33	"key": "authorization",
34	"value": ""
35	}
36	],
37	"method": "post", // the http method to use when calling the function
38	"parameters": {
39	"type": "object",
40	"properties": {
41	"item": { // the name of the input property
42	"type": "string", // the type of the input
43	"description":"" // the description of the input so the agent understands what it is
44	}
45	},
46	"required": ["item"] // the list of required input properties for this function to be called
47	}
48	}
49	]
50	},
51	"speak": {
52	"model": "" // default 'aura-asteria-en' for other providers see TTS Models documentation
53	}
54	},
55	"context": {
56	"messages": [], // LLM message history (e.g. to restore existing conversation if websocket connection breaks)
57	"replay": false // whether to replay the last message, if it is an assistant message
58	}
59	}