MYA

A voice for Claude Code, as a mod (a plugin of function hooks): talk to your terminal, hear it answer, and see a phone ring when it is your turn.

• Tap Right Ctrl → speak → tap again: your words are transcribed and sent to Claude.
• Right Ctrl + Space → speak → tap: your words are transcribed, translated (default: into English) and sent.
• A one-line MYA band above the prompt shows what it is doing: listening (with a moving level meter), thinking, speaking, and, when Claude has finished and waits for you, a ringing telephone (with an optional old-fashioned bell: ringSound).
• MYA has a few lines of its own: a greeting when the session starts and a line when it did not hear you (phrases).
• When you sent the question by voice, MYA reads Claude's answer aloud with a human-like Cartesia voice (Katie by default). Long answers are first condensed into a few spoken sentences; code and paths are not read out. Three options for other tastes: an ffmpeg robot effect (voiceStyle: robot), a local formant synthesizer with no key and no network (voiceEngine: machine, eSpeak NG), or a free local neural voice (voiceEngine: piper, see below).
• Hotkey on/off: /mya off and /mya on, or Right Ctrl + Right Shift, switch the hotkey so it never fires while you copy-paste elsewhere (the band turns red and says so). A Ctrl held a while, or used with a key, a mouse click or the scroll wheel (Ctrl+C, Ctrl+click...), is never taken for a tap.
• /mya fills the prompt instead of sending (/mya go sends, /mya stop ends a recording, /mya mute toggles speech).

Requirements

• Linux with an X11 session (not Wayland, macOS or Windows: the hotkey is read with xinput).
• arecord and aplay (alsa-utils), xinput, ffmpeg (only for the optional robot effect and the optional bell), libespeak-ng (only for the optional machine voice; it ships with speech-dispatcher), Python 3 (standard library only, nothing to pip install).
• A Deepgram API key for speech-to-text. A Cartesia API key to hear replies (not needed with voiceEngine: machine).

Install

claude --plugin-dir mods/mya

Then set your keys, either in the plugin's options (/config, stored in secure storage) or in your shell environment (DEEPGRAM_API_KEY, CARTESIA_API_KEY). Never put keys in a file you commit. Set language to what you speak (en, fr, es, de...) and translateTo to the language you want for Right Ctrl + Space.

Options

| Option | Default | Meaning | |---|---|---| | language | en | Deepgram language code of what you say; also the speech language of replies | | keyterms | Mya | Comma-separated names Deepgram should favour (and that common mishearings are corrected to) | | translateTo | English | Target language of the translate chord | | speakReplies | true | Read answers aloud when the question was voiced | | voiceEngine | cartesia | cartesia (human-like cloud voice), machine (eSpeak NG, local) or piper (local neural voice) | | piperModel | — | Piper engine only: path to a voice .onnx file (its .onnx.json beside it) | | machineVoice | en-us+klatt4 | Machine engine only: eSpeak NG voice and variant (fr+klatt, en+m3...) | | voiceStyle | plain | Cartesia only: plain, robot (buzzing 90s computer) or soft (grit only) | | machineRate / machineWordGap | 155 / 4 | Rhythm of the machine voice: words per minute and the pause between words | | ringSound | false | Play a telephone bell when it is your turn | | phrases | true | MYA's own spoken lines | | voiceSpeed | 1.0 | Speaking speed from 0.5 (slow) to 2.0 (fast); lower it if a voice talks too fast | | voiceId | (empty) | Cartesia voice id; empty = MYA's original voice (Katie) | | showBand | true | Show the MYA band above the prompt | | micDevice | auto | ALSA capture device (auto prefers a USB microphone; see arecord -l) | | speakerDevice | default | ALSA playback device | | python | python3 | Python 3 command for the helpers | | toggleKeycode / chordKeycode / lockKeycode | 105 / 65 / 62 | X11 keycodes: Right Ctrl, Space (translate), Right Shift (hotkey on/off); list yours with xmodmap -pke | | maxTapSeconds | 0.8 | A dictation key held longer than this is not a tap | | hotkeyOnStart | true | Whether the hotkey is active when the session starts |

What it sends where (read this)

• Your microphone audio goes to Deepgram while you record.
• The text of Claude's answer (or its condensed version) goes to Cartesia when replies are spoken. With voiceEngine: machine nothing is sent for speech.
• Translation and condensing use a small haiku call through your Claude Code session. The robot effect runs locally with ffmpeg.
• The hotkey helper reads every X11 key and mouse-button event with xinput test-xi2 --root and keeps only Right Ctrl and its two chord keys; everything else is dropped on the spot, nothing is stored or written. Read bin/hotkey.py (about 60 lines) before enabling it.

Known limits

• The chord's Space also reaches the terminal as Ctrl+Space.
• Recording is manual: it ends when you tap the key again (or after three minutes).
• X11 only. Not tested on other desktops or distributions.

Tests

claude plugin validate mods/mya
claude plugin test mods/mya

Optional: a free local voice with Piper

No account, no credit, no network. Piper is a separate project under the GPL-3.0 licence; MYA does not bundle it, you install it yourself, for the Python you give MYA in the python option:

python -m pip install piper-tts                               # installs piper-tts and pathvalidate
python -m piper.download_voices --download-dir ~/.local/share/piper fr_FR-siwis-medium

Then set voiceEngine: piper, piperModel: ~/.local/share/piper/fr_FR-siwis-medium.onnx (full path) and python to that interpreter. Sentences are synthesized one after the other and streamed into a single aplay, so playback does not stop between sentences as long as synthesis is faster than speech (about 5x on a modest CPU in our test).