Configure third‑party speech‑to‑text (STT) providers - Amazon Connect

Configure third‑party speech‑to‑text (STT) providers

Use the following instructions to configure a third-party speech-to-text (STT) provider.

Prerequisites

Step 1: Open the speech model configuration panel

  1. Sign in to the Amazon Connect admin website.

  2. Choose Bots, then choose the bot.

  3. Choose the locale.

  4. In the Speech model section, choose Edit to open the configuration modal.

    The configuration page for your conversational AI bot.

Step 2: Choose the model type

In the Model type dropdown, choose Speech‑to‑Text (STT). This ensures the locale is configured for transcription rather than speech‑to‑speech.

The speech model dialog box.

Step 3: Review the default speech model settings

By default, Amazon is selected as the speech‑to‑text provider. Review the current settings before switching to a third‑party provider.

The speech model dialog box with Amazon selected as the voice provider.

Step 4: Choose a third‑party STT provider

Open the Voice provider dropdown and choose a supported third‑party speech‑to‑text provider.

The speech model dialog box with Deepgram selected as the voice provider.

Step 5: Enter the model ID and Secrets Manager ARN

  1. In Model ID, enter the provider's model name.

    • Some providers require a minimum or maximum length.

    • Model IDs are case‑sensitive and must match provider documentation.

  2. In Secrets Manager ARN, enter the ARN of the secret that contains the provider API key.

    • The secret must be in the same Region as your Amazon Connect instance.

    • Secrets Manager and KMS key policies must permit Amazon Connect to access and decrypt the key. For more information, see Managing secrets and resource policies.

  3. Choose Continue to save your changes.

Build and activate the locale

If the locale shows Unbuilt changes, choose Build language. The new STT settings become active after a successful build.

Runtime behavior (STT)

  • Amazon Connect routes audio to the chosen third‑party speech‑to‑text provider.

  • No changes to flows or Lambda functions are required.

  • Errors such as invalid credentials or invalid model IDs appear in logs.

  • Metrics and analytics continue to function normally.

Troubleshooting (STT)

  • Invalid model ID: Confirm the value with provider documentation.

  • Access denied: Verify Secrets Manager and KMS permissions.

  • Locale build fails: Ensure required fields are valid.

  • High latency: Validate the provider region configuration.