Dictation and read-aloud
Talk instead of typing, transcribe audio files, hear replies read aloud, get voice notes, and send voice messages from Telegram, Slack or Discord.
You don't need a call to use your voice with Murage. You can dictate a message, turn an audio file into text, have replies read to you, and trade voice messages with your Chief of Staff from your phone's chat apps.
Dictation on a Mac
The microphone in the message box uses Apple's on-device speech recognition. It's free, and the audio stays on your computer. Your words appear as you speak.
- With the message box empty, choose the microphone (Dictate).
- Speak. The words go into the message box.
- Choose the microphone again, or press Esc, to stop. Edit the text if you like, then send.
The first time, allow Microphone and Speech Recognition for Murage.
Voice typing on Windows, Linux and your phone
Where Apple's recognition isn't available, the message box has a Hold to talk button instead.
- Press and hold Hold to talk (or hold Space or Enter while it's focused).
- Speak.
- Let go. Murage sends the recording to be transcribed and adds the text to your draft. Anything you already typed is kept.
Each recording can be up to two minutes. Transcription goes through Flux Router, or your own Groq or OpenAI key if you don't use Flux.
On your phone, voice typing needs a secure link. Turn on Serve on my tailnet on your computer, then open Murage on the https link. See Use Murage on your phone.
Transcribe an audio file
Attach an audio file to a message and Murage offers to transcribe it.
- Attach the file. Murage accepts Ogg, WebM, M4A/MP4, MP3, WAV or FLAC, up to 4 MiB.
- Choose Transcribe with Flux (uses credits). The audio goes to Flux and its transcription provider.
- Check the text under Review transcript and fix anything that's wrong.
- Choose Add transcript to draft, or Discard transcript. Nothing is sent until you send the message.
This needs a Flux Router key.
Read replies aloud
- One message: choose Read this aloud on a bot's reply. Choose it again (Stop speaking) to stop.
- Every reply from a bot: open the bot's settings, go to Voice & alerts, and turn on Read replies aloud. The bot speaks its answers as they arrive, even when you're in another chat.
Replies are spoken in the bot's own voice. Code, file paths and tables are described rather than read out symbol by symbol. The bot needs a voice first. See Voice providers.
Voice notes from a bot
Ask a bot to "send me a voice note", or it may send one when you'd rather hear the answer. It arrives as an audio message in the bot's own voice.
- A voice note is up to 1,500 characters of speech, about a minute and a half. Longer answers are summed up, with the rest in the chat.
- A bot can send up to 3 voice notes in one turn.
- Each note is saved to the bot's files, named "Voice note from name" unless the bot gives it a title.
- When you're talking to your Chief of Staff from Telegram, Slack or Discord, its voice notes are sent there too.
Voice notes need an engine that can use Murage's tools, and a voice for the bot.
Voice messages in Telegram, Slack and Discord
Send your Chief of Staff a voice message in the Telegram, Slack or Discord chat you paired, the same way you would to a person. It understands what you said and replies with a voice message in its own voice. If the answer is long, the detail comes as text too.
Murage must be running on your computer. See Channels.
Troubleshooting
- "Dictation needs Microphone + Speech Recognition access — System Settings → Privacy & Security." Turn both on for Murage, then try again.
- "Dictation isn't available in this build." Use Hold to talk instead, or the installed desktop app.
- "The previous dictation session is still closing. Try again in a moment." Wait a second.
- "Add a Flux key in Settings on the computer to turn on voice typing." Nothing can transcribe yet. Add Flux Router, or a Groq or OpenAI key under Settings → Models.
- "Voice typing needs a secure address." On your phone, use the https link from Serve on my tailnet.
- "Voice typing needs a paid Flux plan. The key is fine; the plan doesn't cover it yet." Your Flux plan doesn't include voice.
- "That was too long. Try again in shorter bursts." Keep each recording under two minutes.
- "Nothing was said in that recording." The microphone picked up silence. Check the right microphone is selected.
- "Audio transcription accepts files up to 4 MiB. Choose a shorter recording." Trim the file or save it in a smaller format.
- "Pick a voice in this bot's settings to read messages aloud". Choose a voice in Voice & alerts.
- "Keep a voice note under 1500 characters..." The bot tried to say too much at once. Ask for a shorter note.