Tachyon Transcripts

Tachyon Transcripts is a fully local, Windows-only meeting transcription application that captures both microphone and system audio in real time, transcribes speech using OpenAI Whisper, and displays live closed captions without sending any data to the cloud.

The tool uses WASAPI loopback to capture dual audio streams — your microphone and any system output — keeping them as separate channels for reliable speaker labelling and clean post-processing. It features a transparent, always-on-top caption overlay that can be toggled with a global hotkey, and automatically selects the optimal Whisper model based on your GPU (large-v3 on high-end NVIDIA cards, medium on smaller GPUs, or a distilled CPU-friendly model as fallback).

After recording, you can re-transcribe sessions offline with higher accuracy settings including beam search, voice activity detection, crosstalk suppression, and speaker diarization using SpeechBrain, pyannote, or Resemblyzer. All transcripts are saved as timestamped Markdown files alongside the original audio recordings.

Everything stays on your machine — no telemetry, no analytics, no cloud services. The only outbound request is the first-time model download from HuggingFace, which can be pre-cached during installation.