# On-device spoken replies Open **Settings → Agent → Agent voice → Where replies are spoken** and choose **On this computer**. Nock then synthesizes spoken replies locally instead of sending reply text to a cloud voice provider. The agent answer itself may still have been produced by a hosted model; this setting controls only speech synthesis. ## Download and select a voice The local voice models are downloaded separately from the installer. The settings card shows download status and available controls. Wait for the model to be ready, then use the local voice picker and its preview button. Six natural voices are available: **Hazel**, **Iris**, **Lark**, **Heather**, **Birch**, and **Alder**. A simpler local voice or your system voice may stand in while the natural model downloads or if the computer cannot run it fast enough. Local speech is free on every plan. On-device speech currently targets English. For another language, Nock may use the system voice instead. The app checks whether the computer can keep up; when the natural model is unavailable or too slow, it uses a supported fallback rather than claiming the model is active. ## Understand what stays local The selected local voice does not upload the reply text for synthesis. That is separate from the answer-generation route: a hosted agent request can still send your prompt and selected context to its configured model provider. Other optional account, update, and connected-app traffic is also controlled separately. See [privacy controls](/nock/privacy-controls). ## When spoken replies are silent 1. Confirm **Agent voice** is enabled. 2. Check **Where replies are spoken** and wait for any local download to finish. 3. Use the preview control, then check system output volume and the selected output device. 4. Read the status in settings. It may explain that the device is unsupported, a download is incomplete, or a system voice is being used instead. An answer can appear on screen even when speech is disabled or unavailable. Nock may speak a shorter line than the full visible answer. See [voice settings](/nock/voice-settings) and [voice troubleshooting](/nock/voice-troubleshooting).