local-transcriber

local-transcriber is a free, open-source app for turning audio and video recordings into text on your own computer. It's aimed at people handling private interviews, client calls or long talks who want transcription without uploading recordings to a cloud service. It runs fully offline once the required models are downloaded.

The browser interface combines editable transcripts with synchronized playback. Words highlight as the recording plays, and selecting a transcript line jumps to that point in the audio. You can search within a transcript, copy text with or without timestamps, and process several recordings in a batch. A local library saves completed transcripts and lets you search across past recordings.

Optional speaker labeling uses sherpa-onnx to distinguish speakers and color-code their lines. It doesn't require an account or a Hugging Face token. For summaries, clickable chapters and action items, the app connects to a local Ollama model, with qwen3:8b as its default. Transcription still works without Ollama.

Whisper runs through faster-whisper, with CPU processing and optional NVIDIA GPU acceleration. The app supports Windows directly and other platforms through Docker. Teams with larger queues can use a self-hosted service with Kubernetes and KEDA to scale transcription workers according to demand.

The project uses the MIT license. Exports include plain text, SRT subtitles, VTT captions, Markdown, timestamped JSON and Word documents. The app removes uploaded source files after processing and keeps transcripts locally.