Products Offline speech recognition

Speech to text in 1,600+ languages, without your audio ever leaving the building.

MetaVox is a fully offline speech recognition platform. It identifies the language being spoken, sends the recording to the model best suited to that language, and returns an accurate, timed transcript. Everything runs on your own GPU hardware, inside your own network, with no connection to the internet.

Request a demo See how it works
The MetaVox interface, with Transcribe, Batch and System tabs, an audio drop zone and automatic language detection
Languages transcribed
1,600+
Faster than real time, in every language
10x+
Speakers identified per recording
Up to 8
Internet connections needed
0

Speeds are measured on NVIDIA DGX Spark, the most compact system MetaVox runs on. Server-grade hardware is faster still.

Built for the people doing the work

Drop in a recording, read the transcript

Upload an audio file and MetaVox does the rest. The spoken language is identified before the real work begins, so it can be corrected up front, and the transcript comes back with clear labels showing what was detected and how certain that was.

  • Automatic language identification, with an override when you know better
  • Optional speaker labels showing who said each passage
  • Timestamps, with text and subtitle-ready exports
  • Copy the transcript or download it in the format you need
A finished MetaVox transcript of a two-person meeting, labelled by speaker, with detected-language, confidence, duration and word-count chips

Whole folders, with the wait stated up front

Point MetaVox at a folder on a connected drive and it identifies the language of every file first, then tells you how long the whole run will take before any work begins. A single long recording works the same way: it is never refused, and the estimate comes first. Transcripts are saved next to their recordings.

  • Drives are found automatically when they are plugged in
  • Files with an uncertain language are set aside for a person to confirm, not guessed at
  • Existing files are never overwritten
  • Batch access is a separate permission, and each batch is visible only to its owner
MetaVox estimates: a long recording that will take a few minutes, with its language editable, and a batch of files ready to transcribe

Know exactly what is installed

The System view shows what the installation can transcribe, one card for each route with the languages it covers, and keeps a full record of the software behind each one, ready for an assurance review.

  • Coverage set out by language
  • A full record of every installed component
  • A live view of how busy the system is
MetaVox System view with headline figures and coverage cards showing the languages each route handles

Strong results in every language, with no effort

MetaVox does the hard work behind the scenes, so your teams get accurate transcripts without needing to know how.

Automatic

The right model, every time

MetaVox identifies the language in each recording and picks the most accurate model for it. Nobody has to choose.

Specialists

Specialist accuracy where it counts

Widely spoken languages are handled by models built specifically for them, for the strongest results in the languages you use most.

No gaps

Nothing turned away

Less common languages and dialects still get a transcript, so a mixed caseload needs only one tool.

Who said what

Speaker labels

On request, every passage is labelled with who spoke it, in any language.

Fair share

Built for teams

Several people can work at once, and one person's large batch never holds up anyone else.

API

Connects to your systems

Everything in the interface is also available to your own tools and workflows through a secure API.

Designed for audio that must stay private

Recorded speech is often personal and sometimes highly sensitive. MetaVox is built to run on a closed network, and its controls are mapped to NCSC guidance.

  • Sign-in for every user, and a secure token for every script
  • Encrypted connections through a single entry point
  • No route to the internet from any part of the system
  • Hardened, locked-down services
  • Transcripts are never stored, and audio is deleted after each job
  • An audit trail of who did what, without recording what was said
  • Every component from a verified source
  • Fair sharing, so one person's large batch never holds up anyone else

Runs on the hardware you choose

  1. Start small with DGX Spark

    MetaVox is optimised for NVIDIA DGX Spark, a compact desk-side system that is all a team needs to get started.

  2. Scale up to the data centre

    It runs just as well on DGX workstations and data-centre GPU servers such as the NVIDIA B300, provided there is enough GPU memory, and gets faster with every step up. We can help you choose.

  3. Installed offline

    Prepared once on a connected machine, then installed on your closed network from a verified package. MetaVox and MetaLex can share a single machine.

See MetaVox on your own audio

Talk to us about a demonstration, a pilot on your hardware, or help choosing the right hardware for your workload.

Request a demo
Also from Devsultants MetaLex: private translation. Translate text and documents on your own hardware, with nothing leaving the machine.