Agent voices and custom voices

  • Desktop, mobile and web
  • All plans

What it does

Decides how spoken replies sound, in Voice Mode and in Read Aloud. There is a default voice for the account, and any agent can have a voice of its own, so several agents can be told apart by ear. A voice is either an included one or a voice of your own ElevenLabs account, including one created from an audio sample. Any voice can go through effects: pitch, speed, volume, and a robot, vocoder or radio filter.

Where to find it

  • Settings > Composer > Reply voice: the default voice and its effects.
  • Settings > Composer > Custom voices (ElevenLabs): your ElevenLabs key, the voices of that account, Create a voice from a sample.
  • Edit agent > Voice (under Advanced options): Give this agent its own voice, then its voice and effects.
  • Voice Mode panel gear: Own voice for (agent name), then Reply voice, Pitch, Speed, Volume and Effect. With the switch off, these rows change the default voice.
  • On the phone: Edit agent > Voice.

How to use it

  1. Give an agent its own voice: Edit agent > Voice, turn on Give this agent its own voice. It starts as a copy of the default voice.
  2. Pick the voice and the effects, and press the play button next to the voice to hear them together. Nothing is billed for the effects.
  3. For a mechanical voice, set Effect to Robot or Vocoder. With Vocoder, Pitch moves the note the machine speaks on.
  4. For a voice of your ElevenLabs account: paste your key in Settings > Composer > Custom voices (ElevenLabs). Its voices appear in every voice picker, and Other voice ID... takes a voice ID directly.
  5. To create a voice from a recording: same card, Create a voice from a sample: a name, one or a few audio files, and the box confirming the voice is yours or that you have permission.

An agent's voice applies to Voice Mode with that agent and to anything read aloud from it (its last answer, its turns in the conversation view, a selection in its terminal). A file preview and anything else take the default voice.

Settings

  • voiceModeVoice (global, synced on the account; Settings > Composer > Reply voice): the default voice. A built-in id (alloy, ash, ballad, coral, echo, fable, onyx, nova, sage, shimmer, verse), or elevenlabs:<voice id>. Default alloy.
  • voiceModeEffects (global, synced on the account; same row): { pitch, speed, filter, volume } on the default voice. pitch in semitones from -12 to 12 (default 0), speed from 0.5 to 2 with the pitch preserved (default 1), filter one of none, robot, vocoder, radio (default none), volume from 0.1 to 1 (default 1, the app's full volume). Settings > Composer > Reply voice > Volume.
  • voiceProfile (agent; Edit agent > Voice, the Voice Mode gear, or agents_save): this agent's voice and effects as one choice, { voice, effects }. Unset = the agent follows the default voice.
  • The ElevenLabs key is not a setting: it is stored encrypted on the account and never comes back to the app, only a mask.

Agent tools (MCP)

  • settings_set with voiceModeVoice / voiceModeEffects: change the default voice.
  • settings_set (scope agent) or agents_save with voiceProfile: give an agent its own voice; null gives it back the default.
  • The key and the samples have no tool: only the person enters them.

Providers

All agent CLIs, no difference: the voice applies to the text being spoken, not to the CLI.

Mobile

On the phone, Edit agent > Voice edits an agent's voice, speed, pitch, volume and effect. Read Aloud reads an agent's last answer with its voice, speed and volume, and everything else with the default voice of the account. Pitch and effects are applied on the computer only: the phone's audio player does not offer them. The ElevenLabs key and the samples are set on the computer.

Limits

  • ElevenLabs bills those replies on your own ElevenLabs plan. AgentsRoom charges nothing for them: no credits and no listening minutes.
  • ElevenLabs blocks its free plan on calls coming from a shared server, which is how AgentsRoom reaches it. Paid plans are not affected, and the card says so when it happens.
  • Creating a voice from a sample needs a paid ElevenLabs plan. At most 5 files and 25 MB. The files go straight to your ElevenLabs account and AgentsRoom does not keep them.
  • When a chosen voice does not answer (key refused, quota spent, voice deleted), an included voice reads the reply and the panel or the status bar chip says so. That reply is billed like any included voice.
  • The vocoder uses about 5% of one CPU core while a reply plays. The other effects are lighter, and nothing runs between replies.

Common questions

  • Can each agent have a different voice? Yes: Edit agent > Voice, or the Voice Mode gear with Own voice for (agent name) on.
  • Can I make spoken replies quieter without lowering my computer's volume? Yes: Settings > Composer > Reply voice > Volume (or the same row in an agent's own voice). It only lowers Voice Mode, Read Aloud and the spoken announcements: music, calls and other apps keep their level. Set it back to 100% for the app's full volume.
  • Can I make a robotic voice? Yes: Effect > Robot or Vocoder, on any voice. It costs nothing.
  • Can I use ElevenLabs? Yes, with your own key: Settings > Composer > Custom voices (ElevenLabs).
  • Can I clone my own voice? Yes, through your ElevenLabs account and a paid ElevenLabs plan: Create a voice from a sample on the same card.
  • Are my voice samples stored by AgentsRoom? No. They are passed to your ElevenLabs account and not kept.
  • Does it use my credits? Not an ElevenLabs voice. Effects never do. An included voice bills as usual.
  • Why did I hear another voice? The chosen voice refused the call, so an included one answered. Settings > Composer > Custom voices (ElevenLabs) says why.
  • Why do effects not play on my phone? The phone applies the voice, the speed and the volume. Pitch and filters need the computer's audio engine.