Voxint

Turn any recording into a speaker-labelled transcript — on your own computer. Voxint transcribes your audio or video, works out who spoke when, and gives you a simple review screen to confirm the speakers and fix the wording before you export.

It is built for individuals and small teams — researchers, journalists, educators — who need their recordings to stay local. Your audio is transcribed on your own hardware by default: no cloud account, no per-minute fees, nothing uploaded. (Two optional features do reach the network when you turn them on: fetching a recording from a URL, and sending transcript text to an outside AI model to polish it. Both are off or opt-in, and clearly labelled.)

Early days (pre-alpha). Voxint works end to end, but it is young software. The interface, database, and settings can still change between releases through the 0.x series. Great for hands-on use and feedback; not yet for mission-critical archives.

See it in action

(All screenshots use a small synthetic three-speaker sample that ships with Voxint.)

Your review queue — completed recordings waiting for your decisions.	Attribute each voice — accept a confident match, judge a heard name, or leave a voice unknown.
Guided setup in the browser — honest readiness checks, plain-language fixes.	A dashboard for run health and throughput at a glance.

What it does

Voxint takes a recording and walks it through four steps:

1. Add your recording — upload it in the browser, paste a URL, or point Voxint at a file it can already see.
2. Voxint does the heavy lifting — it transcribes the words and works out who spoke when, then suggests who each voice is.
3. You review — confirm each speaker, and fix any wording, in a screen built for exactly this. Machine guesses stay separate from your decisions; you always have the final say.
4. Export — download a clean, speaker-labelled transcript (plain text, subtitles, or structured data).

The models all run locally — transcription (Whisper), speaker separation (pyannote), and voice identity (TitaNet). Everything they need is bundled into Voxint, so there is no Hugging Face account or token to set up.

Quickstart

You need Docker with the Compose plugin (≥ 2.24). One command takes a fresh copy to a running console:

git clone https://github.com/bengizmo/voxint.git && cd voxint
./scripts/install.sh

The installer asks only for what it can't invent — an admin password, a folder for your media, and which hardware runs the models — then generates everything else, starts Voxint, waits until it is healthy, and prints the console address.

No graphics card? That's fine. Voxint runs the whole pipeline on an ordinary computer's CPU (needs roughly 8 GB of memory free). It is slower — a long recording can take hours rather than minutes — but it works anywhere. A GPU (NVIDIA, AMD, or an Apple Silicon Mac) just makes it faster.

Have a GPU? How much VRAM you need. The transcription suite (Whisper + pyannote + TitaNet) shares one card and fits comfortably on 8 GB (e.g. RTX 3050/3060 Ti/4060). Turning on the optional bundled local LLM adds ~5 GB, so running everything on one card wants 12 GB (e.g. RTX 3060 12 GB) or more. AMD cards work via the ROCm tier. Full breakdown and card examples → docs/setup.md.

When it finishes, open the console at http://127.0.0.1:8080/ and sign in with the username and password you set. On a fresh install Voxint walks you through a short in-browser setup wizard and an optional guided tutorial on the bundled sample, so you see the whole review loop before pointing it at your own audio.

Full setup for your operating system and hardware → docs/setup.md. First-run walkthrough → docs/onboarding.md.

Using Voxint

Once it is running, these short guides cover the day-to-day tasks:

• Add media & manage runs — upload a file, paste a URL, or watch a folder; follow a run and requeue, cancel, or archive it.
• Review & adjudicate — confirm speakers, correct the transcript, keyboard shortcuts, the waveform, splitting and reassigning segments.
• Manage speakers & export — the speaker roster and the export formats.
• Settings & troubleshooting — configure everything from the browser, and fix common problems.

A little more depth

You don't need any of this to use Voxint, but if you're curious or evaluating it:

• Nothing is lost to a crash. Every run's progress lives in a database, so a restart resumes where it left off, and pausing for review is just a saved row.
• Speaker identity keeps a paper trail. Voxint matches voices against a roster that grows as you use it, and keeps machine proposals strictly separate from your rulings.
• Optional AI polish. Voxint can send transcript text to any OpenAI-compatible model to tidy it up and suggest names — off by default, and a slow or failing model never blocks a run.
• Swappable vocabulary. Names, jargon, and prompts load from a domain pack you can pick per folder, so specialist terms transcribe correctly. See docs/domain-packs.md.
• Measured, not asserted. The models are pinned and their outputs are held to measured-equivalence gates, so an upgrade can't quietly change results. See docs/gpu-contracts.md.

For developers

The console is server-rendered (FastAPI + Jinja + htmx) with a few small React "islands"; the model services are separate containers behind versioned HTTP contracts. To run the code you checked out instead of the release images, layer the build overlays; to work without Docker at all:

uv sync --extra dev
uv run pytest tests/unit
uv run uvicorn voxint.api.app:app --reload

There is also a standalone, database-free scoring harness — pip install voxint gives you the voxint score CLI for speaker-attribution metrics (see examples/). Architecture, contracts, operations, and the release process are documented under docs/.

License

Apache-2.0. See LICENSE and NOTICE. Vendored model weights are redistributed under their own licenses with attribution (titanet: CC-BY-4.0; pyannote segmentation: MIT; WeSpeaker embedding: CC-BY-4.0) — see the provenance files under services/*/models/ and the model-asset releases.