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

# Speech

> How transcription and speech synthesis work behind the scenes

## Transcription

We use a combination of **Deepgram** and **Whisper** for our transcription pipeline. At the moment, we don’t support other STT providers.

Our pipeline includes:

* **Noise cancellation**: Filters background interference
* **Voice isolation**: Focuses on the speaker, even in noisy environments
* **Turn-taking detection**: Helps the agent know when to speak vs. listen

We continuously improve the pipeline, and updates are rolled out automatically.

<Note>
  If you’re experiencing low transcription quality, let us know. We can explore fine-tuning options or support bring-your-own models.
</Note>

## Speech synthesis

We offer six high-quality built-in voices, each professionally cloned and tuned for:

* Natural prosody and intonation
* Accurate pronunciation
* Low-latency streaming

Each voice is replicated across multiple model providers to ensure reliability. If one provider goes down, we can switch seamlessly with no interruption. We recommend starting with a built-in voice for most deployments.

<Note>
  In case you'd like to use a custom voice, we support **ElevenLabs** and **Cartesia**. Get in touch if you’d like to bring your own.
</Note>