Skip to main content

Voice input

Sometimes you want to ask a question while reading a chart on screen, walking, or just because typing is slower than talking. QRY has built-in voice input: click the mic, speak, the transcribed text appears in the prompt as if you'd typed it.

Goal​

You finish this page with at least one prompt sent via voice.

Prerequisites​

  • A working microphone and the browser's mic permission for your tenant URL.
  • The speech-to-text provider configured by your admin in Admin > System Settings > Speech. If the mic icon is disabled or shows a tooltip "Speech provider not configured", ask an admin.

Steps​

1. Click the microphone icon​

In the prompt area, the mic icon sits between the file-attach (📎) and send (▶) buttons. Click it.

The first time, your browser asks for microphone permission. Allow it. The icon turns red while recording.

2. Speak naturally​

Voice input uses VAD (voice-activity detection) — QRY auto-detects when you stop talking and ends the recording. Long pauses end the dictation; short pauses don't.

You don't need to mention punctuation. The transcription model handles natural sentence breaks, but if you want a literal comma you can say "comma" or "new line" explicitly.

3. Review and send​

The transcribed text appears in the prompt. Review for accuracy before sending — voice models have known limitations:

  • Acronyms — MRR may transcribe as MR, empire, or M-R-R. Worth a quick fix before sending.
  • Table or column names — customers_v2 rarely comes through clean. Edit before sending.
  • Names — uncommon names get phonetic spellings.

Press Enter or click Send to submit.

What's happening behind the scenes​

  • Audio is captured by the browser at 16 kHz mono and streamed to the backend.
  • The backend forwards it to the configured provider (cloud or self-hosted).
  • The transcription is returned as text and inserted into the prompt input.
  • The audio is not retained by default — only the transcribed text becomes part of the conversation.

Common issues​

The mic icon is disabled. Speech-to-text isn't configured for your tenant. Ask an admin to set up a provider in Admin > System Settings > Speech.

Browser keeps asking for mic permission. Your browser is set to ask every session. Open the tenant URL's site permissions and grant mic permanently.

Transcription cuts me off mid-sentence. VAD ends recording on long pauses. Speak more continuously, or shorten your prompt.

Wrong language detected. Most providers default to the language set in your tenant's branding. If yours is different, ask an admin to switch the default; some providers also auto-detect on a per-clip basis.

Background noise garbles the transcription. The provider does its best with what it gets. Use a quieter environment or a headset mic.

See also​

QRYA product of IXEN.