Voice input

Voice input turns your speech into text in the message box: record a short message, DataMind Oda writes it out, and you review the text before sending. It is a plugin you switch on, and its speech models need to be prepared before first use; Set up voice input covers that part.

Turn voice input on

Voice input ships switched off.

1. Open Plugins in the sidebar; Plugins and bundles covers that page.
2. Turn on Voice input. Its switch reads Enable Voice input, and the card carries a blue waveform icon.

The first time you record with unprepared models, a setup prompt offers Go to setup and Later; nothing downloads until you start the preparation in the details. Set up voice input walks through it.

Until the models are ready, the microphone stays disabled, and its tooltip reads Prepare recognition in the voice plugin settings first.

Record a message

The microphone sits in the message box, between the model picker and Send message.

1. Click the microphone. Its tooltip reads Dictate; screen readers announce it as Start recording.
2. Allow microphone access if your browser asks. While the browser asks, the toolbar reads Allow microphone access to continue…; while you speak, it reads Recording… and shows a waveform of your voice.
3. Click Stop and transcribe when you are finished. The toolbar reads Transcribing…, or Waking… while the local recognizer starts up.
4. Review the text. Oda places it in the message box at the point where your cursor was when you started recording, as ordinary draft text: it does not send itself, and you can edit it before sending.

• A recording stops by itself at the maximum recording length, 120 seconds by default. Recordings are also capped in size, and a recording over the limit reads The recording exceeds the service limit. Try a shorter recording.
• If your draft changed while you recorded, Oda keeps the transcript for you: the toolbar reads Your draft changed. The transcript is preserved and can be inserted at the current cursor. Choose Insert text to put it in, or Discard transcript to drop it.
• To redo a take, choose Record again.

Cancel or retry a recording

• To stop without keeping anything, press Cancel or Escape while you record. Switching to another window or hiding the tab also ends the recording; nothing is transcribed, and the toolbar reads Voice input cancelled.
• While the browser is still asking for microphone access, switching windows does not cancel the recording; answer the prompt to carry on.
• If Oda heard nothing it could turn into words, the toolbar reads No speech recognized.
• If the recording was cut off before it finished, the toolbar reads Recording was interrupted. Please try again.
• If transcription failed, the toolbar reads Speech recognition failed: <message> with the reason. Try again; if it keeps failing, check the preparation on Set up voice input and include the message when you ask your administrator for help.

Where your audio goes

• Recognition runs on the computer running DataMind Oda; Local means that computer, which can be a different computer from the one whose browser you are using.
• Your recording is never stored or uploaded, and it never reaches the model: it does not appear in the chat, it is not kept as a file or an attachment, and only the text you insert is recorded.
• The microphone needs a secure connection, or a local one, and your operating system's permission. If a recording cannot start, check those first.
• Oda listens only while you record: no always-on microphone, no wake word, no live captions as you speak, and no spoken answers.

Questions, answered

Does dictation send the message for me?

No. The transcript goes into the message box as ordinary text, at the point where your cursor was when you started recording. Nothing is sent until you send it yourself, and you can edit or remove the text first. The message box covers sending.

How long can a recording be?

A recording stops by itself at the maximum recording length, 120 seconds by default. Recordings are also capped in size, and a recording over the limit reads The recording exceeds the service limit. Try a shorter recording. For a long message, record it in a few parts.

How do I cancel a recording?

Press Cancel or Escape while you are recording, or switch to another window or hide the tab; the recording ends without being transcribed, and the toolbar reads Voice input cancelled. If the browser is still asking for microphone access, switching windows does not cancel the recording.

Why does the toolbar say Allow microphone access to continue…?

Oda is waiting for your answer to the browser's microphone permission prompt. Allow access to carry on; if the prompt was dismissed, no recording starts, and you can try again.

What does Your draft changed. The transcript is preserved and can be inserted at the current cursor. mean?

You changed your draft while the recording was being written out, so Oda did not insert the text by itself and kept the transcript for you. Choose Insert text to put it in at the cursor, or Discard transcript to drop it.

What does No speech recognized mean?

The recording did not contain speech that could be turned into text, so there is nothing to insert. Start a new recording when you are ready.

What does Recording was interrupted. Please try again. mean?

The recording was cut off before it finished, so it was not transcribed. Start a new recording; nothing from the interrupted one is used.

Are my recordings saved anywhere?

No. A recording is never kept: it does not appear in the chat, it is not saved as a file or an attachment, and it never reaches the model. Only the text you put in the message box is recorded, and it joins the chat when you send it.

If I switch Voice input off, do I lose the prepared models?

No. Switching the plugin off stops a recording that is running, and the prepared models stay on the computer; switching it back on does not download them again.