Dictate AI prompts, email and notes on Windows

Speak the context for an AI prompt

1. Click the prompt field in a compatible Windows browser or app.
2. Start dictation with your configured shortcut. Explain the task, context and requested output.
3. Finish, review names and details, and submit the prompt yourself.

Talkasaurus transcribes locally. An online AI service receives the text when you use that service; its privacy practices apply. Talkasaurus does not generate an AI answer or submit a prompt automatically.

Draft an email or message

Click the message field, dictate a short draft and review it before sending. Check names, numbers and punctuation carefully. Protected or elevated fields can block insertion, and compatibility varies between applications.

Capture notes while an idea is fresh

Dictate into a compatible note or document field and save the document in that application. Talkasaurus does not keep a transcript-history library. A compatible foot pedal that sends a keyboard key can be assigned as a shortcut; keyboard and supported mouse shortcuts are also configurable.

Listen while reviewing a passage

Select text and use the read-aloud shortcut. The bundled voices run locally. Listening gives you another way to review a draft; selection compatibility varies between editors.

Before you try a workflow

Complete the local sample test in setup first. The beta is free and unsigned, and hardware testing is still limited. Read the product facts and setup guide, or watch the demos.

Supported Windows beta requirements

Windows 10 version 21H2 (build 19044) or newer, or Windows 11, on an x64 PC. English dictation, a microphone, and a compatible Vulkan GPU with a working graphics driver are required. Compatible AMD, NVIDIA and Intel graphics use the shared Vulkan interface; this does not mean every card has been tested. Windows ARM, macOS, Linux, phones and tablets are not supported by this beta.

Setup runs a local speech sample before accepting a model. Performance depends on the GPU, driver and chosen model.