Open chud

Documentation

Voice and audio

Live voice conversation, dictation, read-aloud, and the speech coach.

Live voice conversation#

The orb beside the composer starts a live call rather than a recording: audio streams both ways over a realtime session, so you can interrupt and be interrupted, and the transcript appears as you both speak. A waveform tracks your own microphone level so you can see that you are being heard.

  • A call can act, not just talk. When it generates an image, a chart or a file, that artifact opens in a side panel next to the transcript instead of scrolling away.
  • It can reach the same tools chat reaches — searching, pulling a file in, and the Google exports described on Connections.
  • You can type mid-call, mute the microphone, and switch the voice persona without hanging up.
  • Ending the call leaves the transcript behind as an ordinary chat you can return to.

A voice call is metered like chat, and it is not cheap

Every turn in a call is a real model turn, and speech carries more tokens than the same exchange typed. A long conversation spends the allowance considerably faster than the same conversation in text — that is the honest trade, and the meter is the place to watch it.

Dictation#

The microphone in the composer is dictation, which is a different thing from a call: it transcribes what you say into the input and then stops. Nothing is sent until you press send, so you can edit the transcription first. It uses your browser's speech recognition, so availability and accuracy depend on the browser — Safari and Chrome behave differently, and some browsers do not offer it at all.

Read aloud#

Any answer can be read back from the action row beneath it. Playback is generated speech, so it starts after a short pause rather than instantly, and it can be stopped mid-sentence.

Speech coach#

/m/speech is for rehearsing something you have to say out loud. You paste or write a script, then practise against it: a teleprompter, live pace feedback while you speak, a rhythm trainer, and section-by-section practice for the parts that keep breaking.

  • After a run you get pace over time, a confidence read, and your script marked up where you sped up, slowed down or drifted from the words.
  • Sessions are saved, so you can compare a later attempt against an earlier one.
  • It is a Pro feature.

Written from the code that implements it. Verify against src/lib/m/realtimeVoice.js, src/hooks/useDictation.js, src/lib/m/readAloud.js, src/pages/m/SpeechCoach.jsx.