Skip to main content

Speech-to-text providers

The microphone in the chat prompt lets people dictate instead of typing. Dictation is on by default: with no configuration saved, Qry transcribes with OpenAI Whisper.

How it works​

The browser records the audio — push-to-talk, or hands-free with silence detection that ends the recording on a pause — and sends it to the Qry backend, which passes it to the configured provider and puts the transcription into the prompt. Only the text becomes part of the conversation; the audio is not kept.

Where to configure​

Admin › System configuration › Platform › Speech-to-text.

SettingMeaning
Enable Speech-to-TextSwitch dictation on or off for everyone. Off removes the microphone.
STT ProviderOpenAI Whisper (default) or Google Cloud Speech-to-Text.
Default LanguageAuto-detect (recommended), or a fixed language such as English or Spanish.
Max Recording DurationLongest recording accepted, in seconds (default 60).

Providers​

OpenAI Whisper​

  • Uses the OpenAI API key from System configuration › AI Models.
  • If a Whisper Deployment is set in the Azure AI Foundry settings of AI Models, transcription goes to that Azure resource instead of api.openai.com — authenticated with the resource key, or with the pod's Entra ID identity when no key is set.

Google Cloud Speech-to-Text​

  • Uses the installation's Google Cloud credentials (System configuration › Google Cloud). The service account needs permission to use Speech-to-Text on its project.

Language​

Auto-detect works well for longer phrases. If your teams mix Spanish and English and short dictations come out in the wrong language, set the default to the language most of them use.

Cost​

Both providers charge per minute of audio. Occasional dictation costs little; a team dictating all day scales linearly — look at the first month's usage before assuming.

Common issues​

The transcription is empty or garbled. The browser recorded silence or the wrong input. Check that the right microphone is selected in the operating system.

The wrong language comes back. Short clips are hard to auto-detect. Set a default language.

Transcription fails with 401 or 403. The provider credentials are missing or expired: the OpenAI key, the Azure resource key or role, or the Google service account's permission.

The browser keeps asking for microphone permission. That permission belongs to the user's browser for the Qry address; they can make it permanent in their browser settings.

See also​