Skip to content

Kokoro

Choose Kokoro to hear agents’ replies in a voice made on your own machine, with no account and no key. It has 54 voices in nine languages. Before you add it, install ffmpeg on the machine and allow for a download of about 385 MB. A voice is made on the machine you choose for it, so there is audio only while that machine is online.

Text to speech on this machine: 54 voices in nine languages, free and private. Needs ffmpeg.

  • A voice for agents: an agent can be given one of its voices, and its replies can be listened to.
  • ffmpeg, to read and write audio. Check with ffmpeg -version.
  • A download of about 385 MB the first time, for the models and a speech library.

The model runs on your machine. Nothing is sent to a speech service.

Free. It uses your machine’s own processor and memory.

You can add the module from the machine’s page in the app. To do it by hand, run these on the machine.

  1. Install ffmpeg

    Terminal window
    brew install ffmpeg

    Kokoro uses ffmpeg to read and write audio. Check with ffmpeg -version.

  2. Add Kokoro

    Terminal window
    monoclaw modules add kokoro

    Downloads about 385 MB (models and a speech library) the first time.

  3. Check it

    Terminal window
    monoclaw modules list

    It says Ready once the download is done; the machine’s page in the app shows the same.

  1. Install ffmpeg

    Terminal window
    sudo apt install -y ffmpeg # or your distribution's package manager

    Kokoro uses ffmpeg to read and write audio. Check with ffmpeg -version.

  2. Add Kokoro

    Terminal window
    monoclaw modules add kokoro

    Downloads about 385 MB (models and a speech library) the first time.

  3. Check it

    Terminal window
    monoclaw modules list

    It says Ready once the download is done; the machine’s page in the app shows the same.

  1. Install ffmpeg

    Terminal window
    winget install --id Gyan.FFmpeg -e

    Kokoro uses ffmpeg to read and write audio. Check with ffmpeg -version.

  2. Add Kokoro

    Terminal window
    monoclaw modules add kokoro

    Downloads about 385 MB (models and a speech library) the first time.

  3. Check it

    Terminal window
    monoclaw modules list

    It says Ready once the download is done; the machine’s page in the app shows the same.

You choose these when you give an agent its voice.

Voice. Default: Heart (female).

  • English (US): Alloy (female), Aoede (female), Bella (female), Heart (female), Jessica (female), Kore (female), Nicole (female), Nova (female), River (female), Sarah (female), Sky (female), Adam (male), Echo (male), Eric (male), Fenrir (male), Liam (male), Michael (male), Onyx (male), Puck (male), Santa (male)
  • English (UK): Alice (female), Emma (female), Isabella (female), Lily (female), Daniel (male), Fable (male), George (male), Lewis (male)
  • Spanish: Dora (female), Alex (male), Santa (male)
  • French: Siwis (female)
  • Hindi: Alpha (female), Beta (female), Omega (male), Psi (male)
  • Italian: Sara (female), Nicola (male)
  • Japanese: Alpha (female), Gongitsune (female), Nezumi (female), Tebukuro (female), Kumo (male)
  • Portuguese (Brazil): Dora (female), Alex (male), Santa (male)
  • Chinese: Xiaobei (female), Xiaoni (female), Xiaoxiao (female), Xiaoyi (female), Yunjian (male), Yunxi (male), Yunxia (male), Yunyang (male)

Speed: a number from 0.5 to 2, in steps of 0.05. Default: 1.

Volume (dB): a number from -12 to 12. Default: 0. After levelling: 0 is the standard level.