Professional dictation. Fully offline.
SuperSpeech is built by Wolfy Tech LLC. We build dictation software that respects your privacy and works at lightning speed — no cloud, no compromises.
Why we built SuperSpeech
We were frustrated with dictation tools that either send your audio to the cloud, take seconds to respond, or cost hundreds per year. Doctors, lawyers, and professionals deserve a tool that keeps their confidential data on-device while still being faster than typing.
So we built SuperSpeech: a desktop app that runs state-of-the-art ML models directly on your hardware. No internet connection needed. No data leaving your device. Under one second of latency.
Our principles
Privacy first
All audio processing happens on your device. No cloud uploads, no telemetry, no compromises. Your words belong to you.
Speed
Under 1 second latency on Apple Silicon with Neural Engine. 1-2 seconds on Windows with GPU. Faster than the cloud because there is no network round-trip.
Cross-platform
Native apps for macOS and Windows with platform-specific hardware acceleration. Apple Neural Engine on Mac, CUDA/DirectML on Windows.
Multilingual
25+ European languages in a single model. Automatic language detection. Powered by NVIDIA Parakeet-TDT with 600 million parameters.
Technology
SuperSpeech uses the NVIDIA Parakeet-TDT-0.6B speech recognition model — a modern transformer model with 600 million parameters, trained on 25+ languages. On Mac, we convert the model to Core ML format for native Apple Neural Engine acceleration. On Windows, we use ONNX Runtime with a cascade of execution providers: CUDA, TensorRT, DirectML, and CPU fallback.
We also offer tools for developers: TranscribeCLI for batch transcription from the command line and TranscribeAPI as a self-hosted REST API for integration into custom workflows.