> For clean Markdown of any page, append .md to the page URL. > For a complete documentation index, see https://docs.rydhm.ai/llms.txt. > For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.rydhm.ai/_mcp/server. # Voices The **Voices** page in the sidebar is the voice library: every voice your experts can speak with, from Cartesia, Deepgram, Google and OpenAI. The page header shows how many voices and languages are available right now. Anyone in the workspace can browse and listen; giving an expert a voice is an admin's change. ![The Voices page with filters for provider, language, accent and gender above a grid of voice cards](/_fern-files/_img/fdb30055e65681a62a1a07eef91b4545a1a26f1b745e93e45974c4a93db933f7.webp)![The Voices page with filters for provider, language, accent and gender above a grid of voice cards](/_fern-files/_img/f439eba0b2667e0fe313f5f8d564b21fd91510369c071c7d146de4c8b17791bc.webp) ## Find a voice #### Narrow the list Use the controls above the grid: * **All voices**, **Saved** or **In use** (voices one of your experts speaks with). * **Search name, style, accent or voice ID…** * **All providers**, or one provider. * **All languages**, or one language. With a language picked, **Native speakers** keeps only voices whose first language it is. * **Any accent**, or one accent. * **Any**, **Female** or **Male**. * Sort by **Recommended** or **Name A–Z**. Every dropdown shows how many voices each choice would leave. **Clear** resets the filters. #### Listen Press the orb on a card to play the voice. With a language picked, it speaks that language when the voice can. Press it again to stop. #### Shortlist Press the bookmark on a card to add it to **Saved** while you compare. Saved voices are kept in this browser only. Each card shows the voice's name, provider, language and accent, gender, how many other languages it speaks, its voice ID, and which of your experts already use it. ## Give a voice to an expert There are two ways, and both end in the same place. #### From the Voices page 1. On a card, choose **Use voice**. 2. In **Use** *voice*, pick the expert that should speak with it. 3. Choose **Give** *expert* **this voice**. The change is saved at once. A toast offers **Open expert** to see it on the Voice tab. #### From the expert's Voice tab 1. Open the expert, then the **Voice** tab. 2. Under **Voice**, the current voice is shown as a card you can play. Choose **Change voice**. 3. In **Choose a voice**, press a voice to hear it, then **Choose**. 4. Choose **Save changes**. ![The Choose a voice dialog opened from the Voice tab](/_fern-files/_img/ed64e20bc819ec58e509786c8172e105fd219dc601fae4b2e3a61e539e601e09.webp)![The Choose a voice dialog opened from the Voice tab](/_fern-files/_img/afd07c814d13a19620a9e8d62e69da6ec0629d8459501230e267440a1c07536f.webp) The picker only shows voices the expert's engine can speak with: Cartesia and Deepgram for Standard (the **Balanced** style), Google for Gemini Live (**Fastest**), OpenAI for GPT-Live (**Most natural**). If the expert speaks one language other than English, the picker opens filtered to it. > **Warning** > > **The voice decides the engine.** From the Voices page, a Google voice moves an expert to [Gemini Live](/experts/voice-engines#gemini-live) and an OpenAI voice moves it to [GPT-Live](/experts/voice-engines#gpt-live); a Cartesia or Deepgram voice moves it to [Standard](/experts/voice-engines#standard). The dialog warns you before it saves. The expert's other settings are kept, so switching back restores them. A **GPT-Live** expert speaks with two voices: its own in one-on-one conversations, and a Standard voice in meetings, set under **In meetings** → **Meeting voice**. Both count as **In use**. ## Make the voice sound right The **Voice controls** section on the Voice tab tunes how the chosen voice delivers each line: * **Speed**: from slower to faster (0.8× to 1.2×). Cartesia nudges its pace, Deepgram follows it exactly, and Gemini Live and GPT-Live are asked to. * **Tone**: **Neutral**, **Calm**, **Warm** or **Upbeat**. Exact on Cartesia voices in English; other engines follow it loosely. Not set, the voice keeps its own delivery. * **Pronunciations**: words the voice gets wrong, and how to say them. Choose **Add a word**, then fill **Term** and **Say it like**, spelled the way you'd say it (for example *Nginx* → *engine X*). Tick **Match case** to match only that exact capitalization. Only whole words match. Up to 100 words. * **Backup voice**: speaks only if the main voice's provider fails before a reply starts. It comes from the other provider, so one outage can't silence both. Not available on Gemini Live. * **Deep thinker**: for questions a quick answer could get wrong, the voice model first asks a stronger model (**GPT-6.1 Terra** or **Claude Sonnet 5**). Those answers take 2–4 s longer. Off by default; not offered on GPT-Live. * **Thinking sound**: a soft keyboard sound while a tool keeps the caller waiting for more than a second. Calls only; meetings stay quiet. Choose **Hear the first message** to check the result. The preview applies your speed, tone and pronunciations. ## Gotchas * **A short list may be a provider outage.** If a provider's full catalog can't be loaded, its recommended voices stand in and the page says so. * **Some voices are made for one language.** Check the card's language line before giving a voice to an expert that speaks something else. See [Languages](/experts/languages). * **Gemini and GPT-Live previews take a few seconds** to start. ## Related * [Voice engines](/experts/voice-engines) * [Languages](/experts/languages) * [Provider keys](/workspace/provider-keys)