Skip to content

Voice input

Voice input lets you dictate a message instead of typing it. Intra AI transcribes your recording and puts the text in the message box. Nothing is sent until you select Send.

You need a microphone and a browser that can record audio. Trial accounts don't include voice input.

Find the microphone

The Voice input button (a microphone icon) is in the row below the message box, to the right of Md. If you don't see it, one of these applies:

  • Voice input is not set up on your instance, or it has been turned off. Ask your administrator.
  • Your browser can't record audio. Open Intra AI in a current browser.
  • You are using a trial account.
  • The agent this conversation uses is no longer available, so the message box is locked.

Dictate a message

  1. Below the message box, select Voice input. This works in a new or existing conversation and in a project. The recording bar takes the place of the message box.
  2. If the browser asks for microphone access, allow it. While it waits, the bar reads Requesting microphone access ….
  3. Start speaking when Recording appears. A timer and a level meter run while you talk.
  4. When you're done, select the check mark Use recording. The bar shows Transcribing…, then the text appears in the message box. If the box already had text, the new text is added after it, separated by a space.
  5. Check names, numbers and technical terms, fix anything that came out wrong, and send.

To start over, select the X Discard recording. That recording is not transcribed. Once Transcribing… is showing, you can no longer discard it.

You can record several times in a row. Each transcript is added to the end of what's already there. For editing and sending, see Writing messages.

How long a recording can be

A single recording can run just under ten minutes. At 9:55, Intra AI stops it and transcribes it right away, the same as if you had selected Use recording.

Note: If transcription fails, the recording is gone. For longer texts, several short recordings are safer than one long one.

The platform also sets these limits:

  • By default, a recording can be at most 25 MB. A recording of just under ten minutes normally stays well below that.
  • By default, each person can make 30 transcription attempts in 5 minutes. Failed attempts count too.
  • Speech recognition capacity is shared with other users. When it's busy, you get a message and can try again a moment later.

Language

You can't choose the language, and Intra AI doesn't detect it from the recording. Unless a language has been set for your instance, speech recognition assumes German. If you dictate in English or another language, read the text carefully before you send it.

You'll get better results speaking close to the microphone in a quiet room. Unusual names are usually quicker to fix with the keyboard afterward.

What happens to your recording

Intra AI sends the recording to an external, Germany-hosted speech recognition service based on Whisper. Intra AI itself keeps the audio in memory only while this happens. It does not write the file to disk or store it in the database. What comes back is the recognized text.

The text only becomes part of the conversation when you send the message. If you delete it before that, it isn't kept in the conversation.

Each successful transcription counts as 25 tokens per second of audio and shows up under Token activity in your Profile. It does not reduce your token balance. See Profile, usage and token balance, and for how your data is handled in general, Your data and privacy.

Troubleshooting

Message, What to do
MessageWhat to do
"Microphone access denied. Please allow access in your browser settings."Allow the microphone in your browser's site settings for the Intra AI address, then reload the page.
"No microphone found."Connect a microphone or turn it on in your operating system.
"Could not start recording."Check whether another app is using the microphone, then try again.
"No audio from the microphone. …" or, with your device's name, "No audio from '…'. …" above the recording barThe device hasn't sent any signal at all since the recording started (checked from the 4-second mark). This often happens with virtual audio devices. The recording keeps running. Discard it, pick a different microphone in your browser settings and record again.
"No audio captured."Record again.
"No speech detected."Speak closer to the microphone and a little louder.
"Speech recognition is busy right now. Wait a moment and try again."Wait briefly, then record again.
"Too many requests in a short time. Wait before you try again."You've reached your transcription limit (30 attempts in 5 minutes by default, failed ones included). Wait as long as the message says, or a few minutes if it names no time.
"Transcription failed." or another message such as "Could not decode audio"Record again. If it keeps failing, type the message and let your administrator know.