Skip to main content
PromptQuorumBuilt for humans. Structured for AI.
Home/Power Local LLM
Voice, Speech & Multimodal

TranscriptionSuite Review: Free Local Speech-to-Text with Speaker Labels

·9 min readBy Hans Kuepper · Founder of PromptQuorum · Discovery engine for open-weight & open-source AI

TranscriptionSuite is a free, GPL-3.0 speech-to-text app that runs speech models on your own computer and adds live transcription, speaker labels, a calendar-based audio notebook and an OpenAI-compatible API. Its README says audio and transcripts never leave your machine. This review covers version 1.3.10 and is based on the public website, README and GitHub releases, not on hands-on testing.

Key Takeaways

  • Version reviewed: 1.3.10, released 24 September 2026; the repository was last updated on 4 October 2026, so no stale-project warning applies.
  • License and price: GPL-3.0 or later per the README; free, with no account.
  • Platforms: Windows 11 and Linux (x64) and macOS; the server runs in Docker except on Apple silicon Macs.
  • Standout features: Live Mode, speaker labels, audio notebook, OpenAI-compatible API, webhooks.
  • Maturity: about 775 GitHub stars and a single maintainer who calls the project vibe-coded.

📍 In One Sentence

TranscriptionSuite is an open-source desktop app that transcribes audio with speech models running on your own computer and labels who said what.

💬 In Plain Terms

Think of it as a private transcription studio: record a lecture, dictate into any app or drop in an audio file, and get a transcript with speaker labels without sending the recording to a cloud service.

Read in:
🇺🇸en🇩🇪de🇫🇷fr🇯🇵ja🇨🇳zh🇪🇸es🇧🇷pt🇸🇦ar🇰🇷ko

TranscriptionSuite is a free, open-source desktop app that turns speech into text on your own computer, with live transcription, speaker labels, a searchable audio notebook and a choice of speech models. This review covers what its website, README and GitHub releases document, what they leave out, and how it compares with other transcription tools in the directory.

TranscriptionSuite Review: Free Local Speech-to-Text with Speaker Labels

Get TranscriptionSuite

TranscriptionSuite is a two-part app: a dashboard you install, and a server it downloads and manages for you. The files are attached to the latest GitHub release; the rows below run from easiest to most involved.

Platform
Get It
macOS, Apple silicon (server included)TranscriptionSuite-1.3.10-arm64-mac-metal.dmg (about 3-5 GB, no Docker)
macOS, Apple silicon (dashboard only)TranscriptionSuite-1.3.10-arm64-mac.dmg (about 200 MB; server runs elsewhere)
Windows (x64)TranscriptionSuite.Setup.1.3.10.exe plus Docker Desktop with WSL 2
Linux (x64)TranscriptionSuite-1.3.10.AppImage plus Docker or Podman
Websitetranscriptionsuite.app
Privacy policyNo separate privacy-policy page was found; the README describes local processing
Source codegithub.com/homelab-00/TranscriptionSuite

Only the Mac build with the bundled server is a plain download-and-open install. On Windows and Linux a terminal-level setup of Docker or Podman is required first. The macOS app is unsigned, so macOS blocks it until you clear the quarantine flag or choose Open Anyway; FFmpeg must also be installed. The README lists an Intel Mac build, but the 1.3.10 release shows only Apple silicon disk images.

TranscriptionSuite at a Glance

Attribute
TranscriptionSuite
PlatformWindows, macOS, Linux
PriceFree
LicenseGPL-3.0 or later
RunsLocally; models download on first use
Languages90+ with Whisper; fewer with other models
Real-time transcriptionYes (Live Mode, Whisper models only)
Speaker labelsYes, on recordings and imports; not in Live Mode
Usable on CPU onlyYes; NeMo models are not available
Local API serverYes, OpenAI-compatible transcription endpoint

Attributes follow the speech-to-text comparison used in the Local LLM Software Directory. A missing statement on the site means unknown, not unsupported.

What TranscriptionSuite Is

TranscriptionSuite is a transcription app, not a meeting bot or a cloud service. The README describes an Electron dashboard plus a Python backend that supports Whisper, NVIDIA NeMo, VibeVoice-ASR, SenseVoice, whisper.cpp and MLX models, accelerated by CUDA, Apple Metal or Vulkan, or plain CPU.

The author describes it as a personal tool that grew into a hobby project, built by a mechanical engineer who is not a software engineer; in the README's own words, "this whole thing is vibecoded", though not blindly. The author says they use it themselves and do not plan to abandon it.

This review draws on the website, the README and the GitHub releases. It involved no hands-on testing, so accuracy, speed and stability are not rated here.

How to Get Started

Setup is a dashboard install, then a server start from inside the app. The steps follow the README.

  1. 1
    Prepare your platform
    Why it matters: On Windows and Linux install Docker (Docker Desktop with WSL 2, or Docker or Podman on Linux); on Apple silicon install FFmpeg with Homebrew instead.
  2. 2
    Install the dashboard
    Why it matters: Download the file for your system from the [releases page](https://github.com/homelab-00/TranscriptionSuite/releases/latest); on a Mac, clear the quarantine flag once because the app is unsigned.
  3. 3
    Start the server from the Server tab
    Why it matters: Pick a runtime (CUDA, Metal, Vulkan or CPU) and start it; the first start downloads your chosen model, which can be several gigabytes.
  4. 4
    Record, dictate or import a file
    Why it matters: Use the Session tab to record or import audio, and switch on Speaker Diarization when you need speaker labels.

Features

The feature list is broad for a one-person project. Everything below comes from the README and website.

  • Longform recording. Record from a microphone or system audio and get the full transcript after you stop, with a rolling preview of the latest 20 seconds.
  • Live Mode. Real-time, sentence-by-sentence transcription for dictation, limited to Whisper-family models and without speaker labels.
  • Speaker diarization. "Who said what" labels for Whisper, NeMo, SenseVoice and VibeVoice models.
  • File import. Drop in audio or video and save plain text or .srt / .ass subtitles.
  • Audio Notebook. Recordings land in a calendar view with full-text search and an AI assistant that talks to any OpenAI-compatible provider.
  • Dictation shortcuts. System-wide start and stop shortcuts and paste-at-cursor.
  • Integrations. An OpenAI-compatible /v1/audio/transcriptions endpoint for tools such as Open WebUI and LM Studio, outgoing webhooks, and remote access over Tailscale or a local network.

Models and Hardware

You choose the speech model, and the app detects your hardware and greys out incompatible options. On Docker platforms the choices are WhisperX (faster-whisper models), NVIDIA Parakeet v3 and Canary v2, VibeVoice-ASR, SenseVoice and whisper.cpp; on Apple silicon they are the MLX versions of Whisper, Parakeet, Canary and VibeVoice-ASR.

Language coverage depends on the model: Whisper handles 90+ languages and can translate to English, Parakeet and Canary cover 25 European languages, VibeVoice 51, and SenseVoice five. NeMo models do not run on the CPU runtime, and VibeVoice-ASR is a 9-billion-parameter model of about 16 GB.

The developer claims half an hour of audio is transcribed in under a minute with Whisper on an RTX 3060; this review did not reproduce that figure. The README publishes no minimum RAM or VRAM requirement.

Privacy, License and Cost

The README says everything runs on your computer and that audio and transcripts never leave it. The internet is needed only to download the app and model weights on first use; after that it works offline. Speaker labels with Whisper or NeMo use PyAnnote models, which require a free Hugging Face token.

The AI assistant in the notebook can talk to any OpenAI-compatible provider, including cloud services such as Groq, OpenAI and OpenRouter; anything you send there falls outside the local-only claim. The README also says the app checks for new versions weekly.

The code is GPL-3.0 or later and the app is free. No dedicated privacy-policy page was found. This review did not inspect network traffic, so these statements are the developer's own.

Trade-Offs: Benefits vs. Limitations

Local transcription

What it means in real use:
Audio stays on your machine.
Limitation / caveat:
First-use model downloads are large; no accuracy figures are published.

Speaker labels

What it means in real use:
Know who said what in meetings.
Limitation / caveat:
Not available in Live Mode; Sortformer handles only four speakers.

Many model families

What it means in real use:
Pick speed, language or accuracy.
Limitation / caveat:
Which one fits your hardware is up to you.

OpenAI-compatible API

What it means in real use:
Other tools can use it as a backend.
Limitation / caveat:
The model field is accepted but ignored.

Open source and free

What it means in real use:
GPL code you can inspect.
Limitation / caveat:
One maintainer, self-described as vibe-coded; Docker needed on Windows and Linux.

TranscriptionSuite vs. Alternatives

These are the closest speech-to-text tools in the directory: another meeting transcriber, a Mac dictation app, and two engines you would build on.

TranscriptionSuite

Platforms:
Windows, macOS, Linux
Price and license:
Free, GPL-3.0
Role:
Transcription app with server
Key difference:
Live mode, speaker labels, notebook and API in one

Meetily

Platforms:
macOS, Windows
Price and license:
Free (MIT) and paid Pro
Role:
Meeting transcriber
Key difference:
Built around meetings; macOS and Windows only
Articles about Meetily (1)

Also mentioned in:

MacWhisper

Platforms:
macOS
Price and license:
Freemium, proprietary
Role:
Mac transcription app
Key difference:
Mac-only and closed source
Articles about MacWhisper (2)

Also mentioned in:

faster-whisper

Platforms:
Windows, macOS, Linux
Price and license:
Free, MIT
Role:
Python speech library
Key difference:
An engine for developers, no interface
Articles about faster-whisper (5)

Also mentioned in:

+6 more not shown

Izwi

Platforms:
Windows, macOS, Linux
Price and license:
Free, MIT
Role:
Local voice runtime
Key difference:
Adds speech output and voice cloning behind one API
Articles about Izwi (2)

Platform, price, and feature details for third-party apps change frequently. Verify current specifics on each project's own page before deciding.

Compare in directory

Who Should Use TranscriptionSuite

  • Anyone who transcribes interviews, lectures or meetings. Speaker labels and subtitle export target exactly that.
  • Mac users with Apple silicon. The bundled build is the simplest path and needs no Docker.
  • People who dictate into other apps. Live Mode and paste-at-cursor shortcuts cover it.
  • Tinkerers with a home server. Remote access and the API let other devices use one machine.

Who Should Not Use TranscriptionSuite

  • People who want a no-setup installer on Windows or Linux. Docker comes first.
  • Intel Mac owners. The latest release lists no Intel build.
  • Teams that need vendor support. It is a one-maintainer hobby project.
  • Anyone who needs documented compliance. No privacy-policy page was found.

Frequently Asked Questions

Does TranscriptionSuite need a GPU?

No. A CPU-only runtime exists, but NeMo models are unavailable on it and large models are slow; NVIDIA CUDA, Apple Metal and AMD or Intel Vulkan are optional accelerators.

Why does it need Docker?

The transcription server ships as a container on Windows, Linux and Intel Macs, which the dashboard downloads and manages; Apple silicon Macs run it natively instead.

Can other apps use it as a transcription backend?

Yes. It exposes an OpenAI-compatible transcription endpoint, so tools such as Open WebUI and LM Studio can point at it.

Do I need an account for speaker labels?

Only for the Whisper and NeMo route: PyAnnote needs a free Hugging Face token. SenseVoice, VibeVoice and Sortformer on Apple silicon need none.

Can I run it on one machine and use it from another?

Yes. The README describes remote access over Tailscale or a local network.

Verdict

TranscriptionSuite is a strong fit for people who want private, local transcription with speaker labels and are comfortable with Docker, or who own an Apple silicon Mac where the bundled build removes that step.

The caveats are maturity and transparency: a single self-described vibe-coding maintainer, no privacy-policy page, no accuracy benchmarks beyond the author's own speed claim, and a macOS app that is unsigned. Try it on non-sensitive recordings first.

Sources

← Back to Power Local LLM

MCP