This open-source speech recognition tool converts the human voice in audio and video files into text, running fully offline on your own computer. Built on the faster-whisper model family, it has earned more than 4,800 stars on GitHub and was updated in 2026. Because everything runs locally, it can replace cloud speech recognition APIs where privacy or cost matters.
The tool accepts audio and video uploads through a simple local web interface and transcribes speech with accuracy the author compares to OpenAI’s official API, since it uses the same Whisper model lineage in its faster-whisper implementation. Results can be exported as JSON, as SRT subtitles with timestamps, or as plain text. Several model sizes are available, from the tiny bundled model up to large-v3, trading speed against accuracy and hardware demands, and machines with an NVIDIA GPU and CUDA set up will use GPU acceleration automatically. Prebuilt packages for Windows make it accessible to non-developers, while source deployment supports Linux and macOS as well.
Because recognition happens entirely on the user’s own machine, recordings never leave the device. That makes the tool practical for sensitive material such as legal, medical or internal company audio, and it also removes dependence on network quality: long files can be processed overnight on a workstation without any upload step or usage quota.
The tool is free and open source under the GPL-3.0 license. There are no per-minute transcription fees, because recognition happens entirely on your own hardware.
This tool is for video creators, journalists, students and teams who need subtitles or transcripts without sending recordings to the cloud. It is particularly attractive for transcribing meetings, lectures and interviews in bulk, and for users in regions or organisations where cloud speech APIs are impractical or restricted.