Real-Time Speech-to-Text/AI Models & Voice Engines
Connect Deepgram Nova-3 to IntelDialix voice agents
Speech recognition from Deepgram with background-noise handling for phone-quality audio.
In short
You connect Deepgram Nova-3 to IntelDialix in two directions. When a call starts or ends, IntelDialix sends a signed webhook to Deepgram Nova-3 with the transcript and outcome. During a call, an agent can reach Deepgram Nova-3 through a custom tool — an HTTP request you define — to look up or update information. No SDK is needed; it is plain HTTPS and JSON.
- Deepgram Nova-3
- IntelDialix
- AI voice agent
- Webhooks
- Deepgram Live Streaming WebSocket
- HTTPS
- JSON
How it works
How information moves between a live phone call and Deepgram Nova-3.
- 01
A call reaches your agent
A caller dials a Twilio or SIP number assigned to the agent, or the agent places an outbound call from a campaign.
Twilio / SIP number → IntelDialix agent
- 02
The agent uses Deepgram Nova-3 mid-call
When the conversation needs data, the agent runs a custom tool: an HTTP request to Deepgram Nova-3 with the details it has collected.
Custom tool → Deepgram Live Streaming WebSocket
- 03
The call ends with a record
IntelDialix keeps the recording, transcript and post-call analysis for the conversation in your call history.
Recording · transcript · analysis
- 04
Deepgram Nova-3 receives the result
A signed webhook delivers the call event to your Deepgram Nova-3 endpoint, and is retried automatically if the delivery fails.
HTTPS POST · X-Dialix-Signature
At a glance
Key figures
- Voice engines
- ElevenLabs · Vapi · Gemini
- Choose the engine per agent and compare them on real calls.
- Telephony
- Twilio & SIP
- Use numbers you already have, for inbound and outbound calls.
- Call events
- Signed webhooks
- HMAC-SHA256 signature on every delivery, retried on failure.
- Starter plan
- $14 / month
- One agent, one number and 100 calls a month.
Example
// Verify an IntelDialix webhook before trusting it in Deepgram Nova-3.
// Header: X-Dialix-Signature: t=<unix seconds>,v1=<hex>
// where v1 = HMAC-SHA256(secret, "<t>.<raw request body>")
import crypto from 'node:crypto';
export function verifyDialixWebhook(rawBody, signatureHeader, secret) {
const parts = Object.fromEntries(
signatureHeader.split(',').map((kv) => kv.split('='))
);
const expected = crypto
.createHmac('sha256', secret)
.update(`${parts.t}.${rawBody}`)
.digest('hex');
const valid =
typeof parts.v1 === 'string' &&
parts.v1.length === expected.length &&
crypto.timingSafeEqual(Buffer.from(parts.v1), Buffer.from(expected));
const fresh = Math.abs(Date.now() / 1000 - Number(parts.t)) < 300;
return valid && fresh;
}
// Body: { "event": "call.completed", "payload": { ... }, "timestamp": "..." }Check the signature before acting on an event in Deepgram Nova-3. The secret is the one you set when you add the webhook in IntelDialix settings.
FAQ
Frequently asked questions
How does IntelDialix connect to Deepgram Nova-3?
Through webhooks and custom tools. IntelDialix posts call events to a Deepgram Nova-3 endpoint you choose, and agents can call Deepgram Nova-3 during a conversation using an HTTP tool you configure in the agent editor.
Which events can Deepgram Nova-3 receive?
IntelDialix sends events such as call.initiated, call.inbound, call.completed, call.failed, campaign.started, campaign.completed. Each delivery includes a JSON payload and an X-Dialix-Signature header you can verify with your webhook secret.
Do I need a developer to set this up?
You need an HTTPS endpoint in Deepgram Nova-3 that can receive webhooks — many tools provide one — and, for mid-call actions, an endpoint the agent can call. Building the agent itself needs no code.
Keep exploring
Related pages
- templateDeepgram Nova-3 Multi-Accent Phone AgentAccurate speech recognition handling global accents, background noise, and noisy mobile callers.
- integrationElevenLabs V3 Voice AI IntegrationHyper-realistic synthetic voices with dynamic inflection, expressive stability, and 29-language native accents.
- integrationGroq LPU Voice AI IntegrationLPU-accelerated Llama 3.3 70B inference achieving 500+ tokens/second for instantaneous agent responses.
- integrationOpenAI Realtime Voice AI IntegrationUltra-low latency speech-to-speech interaction powered by OpenAI Realtime models running over native WebSockets.
- integrationClaude 3.7 Sonnet Voice AI IntegrationEmpower phone agents with deep real-time reasoning, tool invocation, and nuanced human-like conversational tact.
- integrationClaude Sonnet 4.5 Voice AI IntegrationCutting-edge reasoning capabilities for complex enterprise support, technical troubleshooting, and multi-turn negotiation.
Put an agent on the phone today.
Create a free account to explore the dashboard. Plans start at $14 a month for one agent and 100 calls.