← All plugins

Speech

by withbuddi · @withbuddi/plugin-speech · GPL-3.0-only

Your team hear you and answer with a voice: on the dashboard and on Telegram. Whisper and Kokoro run on your computer, or a cloud service you choose.

v0.1.4By buddiVoice

What it reaches in buddi

  • It sends web requests.
  • It uses a model account you pick.
  • It reads every file in your Files library.

What leaves your machine

  • api.openai.com: OpenAI accounts: a recording goes to be transcribed, or a text to be spoken; the text or the audio comes back.
  • huggingface.co: Install on Settings → Speech: the Whisper and Kokoro model files, pinned and checked; nothing is sent.
  • *.hf.co: Hugging Face's download servers (us.aws.cdn.hf.co and its like), where huggingface.co sends those files.
  • registry.npmjs.org: Install of Kokoro: the eSpeak NG package (12.5 MB, GPL-3.0), pinned and checked, which pronounces the languages other than English; nothing is sent.

Its tools

Run without asking 2

  • speech.transcribeTurn a recording from the Files library (a voice note, an audio file up to 25 MB) into text. With a cloud service the recording leaves this machine for the listening service the owner chose in Settings → Speech (OpenAI or their OpenAI-compatible server), which sends back the text; with Whisper on this computer nothing leaves. Returns { text, language? }.
  • speech.saySay a text aloud and keep the audio in the owner's Files library (a voice note, OGG/Opus when the service gives it). With a cloud service the text leaves this machine for the speaking service the owner chose in Settings → Speech (OpenAI or their OpenAI-compatible server), which sends back the audio; with Kokoro on this computer nothing leaves, and it speaks English, French, Spanish, Italian, Portuguese and Hindi (the voice follows the text's language). Write for the ear.

Only you, from its own pages 6

  • speech.set_settingsThe Speech settings. The owner's own.
  • speech.testTry the chosen listening service on a short clip. The owner's own.
  • speech.previewPlay a short sample of a speaking service and voice on the Speech page; nothing is kept. The owner's own.
  • speech.installDownload a speech model to run on this computer, in the background. The owner's own.
  • speech.removeDelete a speech model downloaded to this computer. The owner's own.
  • speech.telegram_voiceVoice replies on Telegram: when, and what is sent. The owner's own.

Screens in the dashboard

  • Speech (settings)

What it stores

Its own Postgres schema, speech, inside buddi's private database. Removing the plugin keeps it unless you also drop its data.