TranscriptionSuite
A local speech-to-text app runs offline on Windows, macOS and Linux, with speaker labels, live dictation and CPU or GPU processing.
Alternative to:

TranscriptionSuite is a speech-to-text app for people recording lectures, transcribing conversations or dictating documents on their own hardware. Transcription works offline after the app and model downloads, keeping audio and transcripts local.
- Recording and file import: Capture microphone or system audio with a rolling transcript preview, or transcribe existing audio and video files and export text or SRT and ASS subtitles.
- Live dictation: Get sentence-by-sentence transcription through faster-whisper or whisper.cpp; other model families don't support Live Mode.
- Speaker labels: Identify who said what with supported models, using PyAnnote, built-in labeling in SenseVoice and VibeVoice, or Sortformer on Apple Silicon.
- Model choice and translation: Choose Whisper, NVIDIA NeMo Parakeet and Canary, SenseVoice, VibeVoice-ASR or whisper.cpp, with language coverage and translation capabilities depending on the model.
The app is free and open source under GPL-3.0, with desktop versions for Windows, macOS and Linux. Apple Silicon Macs run MLX models natively through Metal without Docker. Windows, Linux and Intel Macs use a Docker-based server. Processing can use a CPU, NVIDIA CUDA, or AMD and Intel GPUs through Vulkan; model compatibility varies by hardware, and Intel Macs use CPU processing.
PyAnnote speaker labeling requires a Hugging Face account token. SenseVoice and VibeVoice have built-in alternatives that don't need one, while Sortformer provides token-free speaker labeling on Apple Silicon.