> ## Documentation Index
> Fetch the complete documentation index at: https://docs.sentivue.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Deepgram

> Connect Deepgram to SentiVue Talk

Deepgram provides **transcription** of what the caller says, and **voice** for the agent's replies. One connection covers both.

Deepgram is the default transcriber on a new account.

## What you need

| Field   | Required |
| ------- | -------- |
| API key | Yes      |

## Getting your API key

1. Sign in to the [Deepgram Console](https://console.deepgram.com/).
2. Select the project you want the key to belong to.
3. Go to **API Keys** and create a key.
4. Copy it — the key is shown once.

<Warning>
  A Deepgram key belongs to **one project** and to the account that created it. It will not work against a different project, and it stops working if that person is removed from the project or their account is deleted.

  For anything running in production, create the key under an account that will outlast any individual — not a personal login.
</Warning>

Keys can also carry an **expiration date**. A key that expires stops working mid-month with no warning, so either leave the expiry unset or track the date.

## Adding the connection

1. Go to **Connections** and choose **Deepgram**.
2. Paste the API key.
3. Choose which services this connection drives — transcription, voice, or both.
4. Save.

## What you can select afterwards

Choose your models on [Model Configurations](/talk/configurations/inference-providers).

**Transcription model and language** — set the language to match your callers. Coverage differs by model — see Deepgram's [model list](https://developers.deepgram.com/docs/models-languages-overview).

**Voice** — pick from the dropdown, or enter a voice ID manually. The ID carries the model, as in `aura-2-helena-en`, so there is no separate model field. Deepgram's [voice list](https://developers.deepgram.com/docs/tts-models) has samples.

<Note>
  If you need a voice in a language Deepgram does not cover, use a different [voice provider](/talk/configurations/voice) — the transcription can stay on Deepgram.
</Note>

## Using Deepgram for only one service

If you want Deepgram transcription but a different voice, turn the voice service off on the connection.

## Custom hosts

If you reach Deepgram through a proxy or a self-hosted deployment rather than `api.deepgram.com`, you can set a base URL per service on the connection.

## Troubleshooting

<AccordionGroup>
  <Accordion title="The key is rejected">
    Confirm you copied it from the project you intend to use. A Deepgram key cannot be used outside the project it was created in.
  </Accordion>

  <Accordion title="Calls worked and then stopped">
    Check whether the key has an expiration date that has passed, or whether the account that created it has been removed from the project. Either invalidates the key.
  </Accordion>

  <Accordion title="Transcripts are poor or empty">
    Check the language setting first — a mismatch between the configured language and what the caller speaks is the usual cause. If callers switch language mid-call, move from a Flux model to `nova-3-general`.
  </Accordion>

  <Accordion title="A voice ID is rejected">
    Deepgram voice IDs include the model prefix and language suffix, as in `aura-2-thalia-en`. Copy the full ID rather than the voice name.
  </Accordion>
</AccordionGroup>