StarWhisper is a separate Windows app, not an official OpenAI product. It uses OpenAI's Whisper model family through whisper.cpp. Record speech, stop, then review or insert the transcript. Choose local processing after the model is available, or an optional cloud path when you opt in.
OpenAI Whisper is an open-source speech recognition model family released by OpenAI. It is separate from the StarWhisper application. The result depends on the selected model, audio, language, and recording conditions.
whisper.cpp is a C/C++ implementation that can run Whisper without the Python Whisper runtime. StarWhisper uses whisper.cpp for local inference through its Windows interface.
This is the StarWhisper product site and download source. StarWhisper is an independent Windows application, not an official OpenAI product or the upstream whisper.cpp project. The app adds model selection, recording controls, optional preview and streaming settings, and optional cloud transcription around local Whisper models.
Whisper models have different memory needs and response times. Select a model that fits your PC and recording, then use Settings guidance and a short sample instead of treating any score as universal.
| Model | Role in a Windows workflow | StarWhisper Plan |
|---|---|---|
| tiny | Lower-resource local option | Free |
| base | Local model option | Free |
| small | Local model option | Free |
| medium | Larger local model option | Pro |
| large-v3-turbo | Larger local model option | Pro |
| large | Larger local model option | Pro |
Free includes Tiny, Base, and Small, with 500 words per day and 3,500 words per week. Pro is $10/month or $80/year and adds Medium, Large v3 Turbo, and Large plus file transcription. Model files may already be present in the installer or may download through Settings depending on the package; availability, memory use, and response time depend on the selected model and PC.
For dictation, choose a model your hardware can handle and compare a short recording. For saved files, Pro users can choose larger models when they fit the machine. Model size alone does not guarantee a particular result, so review output for the audio and language you use.
The StarWhisper app workflow does not require separate Python, pip, or command-line transcription commands. It packages a Windows GUI around whisper.cpp. Installer setup, model availability, microphone permission, and driver support still depend on the selected package and machine.
Free includes Tiny, Base, and Small. Pro adds Medium, Large v3 Turbo, and Large. Models that are not already present can be downloaded through Settings. The app checks model availability and lets you switch among models supported by your plan and current installation.
StarWhisper can use NVIDIA CUDA when a compatible GPU and driver path are available. It also has an optional Vulkan path for supported NVIDIA, AMD, and Intel hardware when its probe succeeds. If no GPU backend is available, it falls back to CPU inference. Actual speed depends on the model, hardware, drivers, audio, and settings.
Standard dictation records audio and transcribes after you stop. StarWhisper also has opt-in inline preview and streaming settings on Windows. These modes are disabled by default, and their behavior depends on the selected mode, model, hardware, and settings. Review the result before relying on it.
After setup and the required local model download, local mode processes captured audio on your device. Internet may be needed for setup, model downloads, sign-in, periodic Pro license checks, and optional cloud transcription. Cloud mode sends audio to the selected service when enabled and permitted by the app's consent and account settings. See the offline speech to text Windows page for more detail.
OpenAI also offers a hosted Whisper API with its own usage pricing and network requirements. StarWhisper is a separate app that can use local models or an optional cloud path. Compare the path, settings, and data handling that fit your work:
The OpenAI Whisper API documentation is the reference for the hosted API. StarWhisper provides a separate Windows GUI around local Whisper models with an optional cloud path.
Choose Tiny, Base, or Small from the Free plan, then compare a short recording on your PC. Output depends on the model, microphone, language, background noise, and settings. Free usage limits are 500 words/day and 3,500 words/week.
For saved recordings, Pro file transcription supports Medium, Large v3 Turbo, and Large when those models are available. Larger downloads need more disk space and memory and may take longer on CPU. Review the transcript and any subtitle output before relying on it.
Select the spoken language or auto-detect in Settings. Whisper output varies by language, accent, microphone, and recording conditions. Test your own audio and review the result. See the multilingual speech to text guide for setup guidance.
CPU-only inference is supported. Start with a model your machine can handle. Larger models need more memory and can take longer on CPU, so use Settings guidance before downloading one.
StarWhisper is an independent Windows GUI for OpenAI Whisper models
See download optionsOpenAI Whisper is the model family. OpenAI's Whisper API is a hosted service with separate network and usage-pricing terms. StarWhisper is an independent Windows app: local mode processes audio on your device after model setup, while optional cloud mode follows the selected service and app settings.
There is no universal day-to-day choice. Free includes Tiny, Base, and Small. Pro adds Medium, Large v3 Turbo, and Large. Compare a short recording with your own microphone and language because output depends on the full setup.
StarWhisper uses whisper.cpp, so the app workflow does not require Python, pip, or command-line transcription. It provides Windows recording controls, model management, and supported text-field insertion. You still need a working microphone, a model, and a compatible machine.
Local mode processes captured audio on your device after setup and model download. Internet may be needed for setup, model downloads, sign-in, periodic Pro license checks, and optional cloud transcription. Cloud mode sends audio to the selected service when enabled.