Amazon Connect and Deepgram

This guide explains how to integrate your Deepgram models into Amazon Connect flow.

Amazon Connect is AWS’s cloud-based contact center platform, and Amazon Lex provides tools for quickly building conversational chatbots. With Deepgram integrated into Connect, organizations can seamlessly replace AWS Transcribe and Polly with Deepgram’s STT and TTS models—without modifying their existing Connect flows or operational logic.

• Connect/Lex → Deepgram (STT): Live caller audio is streamed to Deepgram for real-time transcription.
• Deepgram → Connect/Lex: The transcript is returned instantly for intent recognition or routing logic.
• Connect/Lex → Deepgram (TTS): Lex sends text to Deepgram TTS model to generate natural spoken responses.

Before starting:

• An active Deepgram account
• An active AWS account
• A Deepgram API key
• An Amazon Connect instance, and Amazon Lex bot if using Lex

1. Sign in to the Deepgram Console.
2. Go to API Keys → Create API Key.
3. Copy the generated key — you’ll need it in AWS Connect.
4. Follow these steps to store the Deepgram API key in your AWS Secrets Manager

If you’re using Amazon Lex bots in your Connect flow, follow the AWS guide to set up Deepgram in Amazon Lex.

Follow the AWS guide to configure Deepgram as the STT provider in your Amazon Connect bots.

1. Under Model Type, select Speech-to-Text
2. Under Voice Provider, select Deepgram
3. Under Model Id, fill in the Deepgram model you want to use, e.g. nova-3-general. Please see the full list of supported languages and models.
4. Under Secrets Manager ARN, use the ARN for you Deepgram API key from Step 1

Similarly, follow the AWS guide to configure Deepgram as the TTS provider in your Amazon Connect bots.

1. In the Set voice block in your Connect flow, select Deepgram in the Voice provider dropdown
2. Under Model, choose Set manually and enter the Deepgram TTS model you want to use, e.g. aura-2.
3. Under Voice, choose Set manually and enter the voice, e.g. thalia. Under Language, choose Set manually and use a language that is supported by the voice, e.g. English. Please see the full list of supported languages and voices. The model and language/voice combination in Connect setting will map to Deepgram’s model name such as aura-2-thalia-en.
4. Under Secrets Manager ARN, use the ARN for you Deepgram API key