StarWhisper
Voice to text reimagined - works offline with local Whisper models
About StarWhisper
StarWhisper is a desktop dictation tool built around a choice most transcription products do not offer: run the model locally or send audio to a cloud API, and decide which based on what you are dictating. Offline mode uses local Whisper models, meaning your voice and the resulting text never leave the machine, with no network round trip and consequently very low latency. Connected mode uses OpenAI for maximum accuracy when the content is not sensitive. For anyone dictating medical notes, legal drafts or anything under an NDA, that first option is not a preference but a requirement.
The application runs on Windows and Mac, with distribution through the Microsoft Store among other channels, and reports over five thousand professional users. Stated performance figures are sub-100ms latency and 98% or better accuracy, with GPU acceleration supported on NVIDIA cards for local inference. Multiple Whisper model sizes are selectable from tiny through large, which is the meaningful trade-off in local speech recognition: larger models are more accurate and slower, and the right choice depends on your hardware and your tolerance for lag. The free tier is genuinely free forever at 500 words per day and requires no account at all, which is unusually generous for a desktop application.
Install the application and choose your mode. For offline operation, select a local Whisper model sized to your hardware, from tiny for older machines up to large where accuracy matters most and a capable GPU is available; NVIDIA acceleration is supported. For maximum accuracy on non-sensitive material, connect it to OpenAI instead. In use it is a dictation layer rather than a separate app you paste out of: speak, and text appears where you are typing, with latency low enough that it does not break the flow of composition. The free tier covers 500 words a day with no account required, so it can be evaluated properly on real work before any decision to pay.
- •Fully Offline Local Whisper - Runs local models with no internet requirement, so audio and transcripts never leave the device
- •Selectable Model Sizes - Choose between tiny and large Whisper models to trade accuracy against speed on your specific hardware
- •Optional OpenAI Connection - Switch to a cloud model when maximum accuracy matters and the content is not sensitive
- •Sub-100ms Latency - Fast enough that dictation keeps pace with thought rather than lagging behind it
- •GPU Acceleration - NVIDIA GPU support for faster local inference
- •Windows and Mac - Native desktop application on both platforms, distributed through the Microsoft Store among other channels
- •No Account Required - The free tier works without registration, email or payment details
- •500 Free Words Daily - A genuinely usable free allowance rather than a trial that expires
For professionals who dictate confidential material and cannot send it to a cloud service: clinicians, lawyers, therapists and anyone working under an NDA, for whom the local mode is the entire reason to choose this over a browser-based alternative. It suits writers and researchers who compose faster by speaking, developers dictating notes and commit messages, and anyone whose hands need a break from typing. Users on older hardware are served by the smaller Whisper models; those with an NVIDIA GPU get the best of the local option. Note it is a desktop application, so mobile dictation is not covered.
