FirstbeamSet up voice input and transcription
Diagnose microphone capture separately from transcription credentials.
First, check microphone capture
Voice input begins with microphone capture in the desktop WebView. Windows microphone access and desktop-app access need to be available. If capture fails before audio is recorded, changing a transcription API key cannot fix that failure.
Use a matching transcription preset
Choose the BYOK transcription provider in Voice & input. Save a matching endpoint, model and key together. Groq transcription uses an OpenAI-compatible audio endpoint; a chat-provider configuration is a separate setting. Changing providers requires checking the stored credential as well as the endpoint.
https://api.groq.com/openai/v1/audio/transcriptions
whisper-large-v3-turboRun a short microphone test
Use the settings microphone test before dictating a long message. Check the selected language and stop the recording to request transcription. Distinguish a capture-permission error from an HTTP authentication, model or endpoint error; they happen at different stages.
Report the stage, not the secret
For support, include the app version, Windows version, capture or provider error category and reproduction steps. Remove API keys and private transcript content. A key should be entered in the dedicated credential setting, never in a project note or shared screenshot.
Read the capture error first
A permission-denied error points to access being rejected before recording. A missing-device error points to available audio input. A device-read error can indicate that the selected device cannot be opened. These are capture-stage categories; an HTTP error after recording belongs to transcription. Use the latest error text and reproduction steps to identify the stage.
A repeatable voice test
Save the provider settings, select the intended dictation language and use a short non-private phrase. Start and stop capture deliberately, then check whether text returns. If it fails, record whether the error appeared before recording, when stopping or after the provider response. This is more useful than changing the key, model and Windows settings simultaneously.