Voice and dictation

Dictation and spoken replies are separate

Speech-to-text turns microphone audio into text. Text-to-speech turns a reply into audio. Configuring one does not configure the other. A build may expose dictation without a complete spoken conversation workflow.

Check which voice features your installed client exposes before adding a service credential.

Check speech input

1. Configure a supported speech-to-text provider on the host.
2. Allow microphone access on the client when requested.
3. Record a short, non-sensitive sentence.
4. Check the transcript before sending it to your coding assistant.

If recording works but transcription fails, check the speech provider’s credentials, endpoint, model, and quota. If no audio is captured, check microphone permission and the selected input device first.

Voice input allowance

Managed voice input has a plan allowance. Free includes five minutes per account each UTC day, Lyre Pro includes sixty minutes, and Lyre AI includes one hundred and twenty minutes, resetting at 00:00 UTC. On every plan a single recording is limited to two minutes, and technical size, rate and concurrency limits apply.

The client shows the remaining allowance and disables capture when it is exhausted until the daily reset.

Audio may leave the device

A hosted transcription service receives audio to process it. Choose the provider with that in mind. Avoid dictating secrets, and correct names or commands before sending a transcript.

See integrations for the distinction between coding, transcription, and other connected services.