homelab-00/TranscriptionSuite

A fully local and private Speech-To-Text app, offering multiple model backends, diarization & calendar mode - Available for Windows, macOS & Linux

What it solves

TranscriptionSuite provides a fully local, private solution for converting speech to text. It eliminates the need to send sensitive audio data to cloud providers by running high-performance transcription and speaker identification models directly on the user's own hardware across Windows, macOS, and Linux.

How it works

The project consists of an Electron-based dashboard (frontend) and a Python-based server (backend). It supports multiple Speech-To-Text (STT) backends including Whisper (via faster-whisper and whisper.cpp), NVIDIA NeMo (Parakeet and Canary), and VibeVoice-ASR.

Depending on the hardware, it utilizes different acceleration paths:

  • NVIDIA GPUs: CUDA acceleration via Docker.
  • Apple Silicon: Native Metal/MLX GPU stack.
  • AMD/Intel GPUs: Vulkan support (via whisper.cpp).
  • CPU: Fallback mode for all platforms.

It also integrates PyAnnote or VibeVoice for speaker diarization (identifying who is speaking) and offers an "Audio Notebook" mode that allows users to chat with their notes using any OpenAI-compatible LLM provider.

Who it’s for

It is designed for users who require high-privacy transcription workflows, such as journalists, researchers, or anyone wanting a local "Audio Notebook" for dictation and meeting notes without relying on cloud services.

Highlights

  • 100% Local: Audio never leaves the machine; internet is only used for initial model downloads.
  • Multi-Backend Support: Compatible with Whisper, NVIDIA NeMo, and VibeVoice-ASR.
  • Speaker Diarization: Identifies different speakers in a recording.
  • Flexible Input: Supports longform recording (mic or system audio), live real-time transcription, and importing existing audio files.
  • Audio Notebook: A calendar-based view with full-text search and an AI assistant for querying notes.
  • Cross-Platform: Native support for Apple Silicon (Metal), NVIDIA (CUDA), and AMD/Intel (Vulkan).