Skip to content

hertz-dev

An open full-duplex audio-only transformer model

Visit official site

What hertz-dev is

The model is described as an open full-duplex audio-only transformer foundation model from Standard Intelligence, aimed at low-latency real-time speech interaction.

What you can do with it

  • Turn mono speech into a low-rate latent representation
  • Predict the next encoded audio frame

Getting started

  • Clone or download the code from the project page
  • Install dependencies, then run encoding and decoding tests

Pros & cons

✓ What we like

  • The model is open and easy for researchers to fine-tune
  • A convolutional audio autoencoder is part of the design

! What to watch out for

  • The described input is mono 16 kHz speech
  • Fine-tuning for each task is done by the researcher

FAQ

How many parameters?

The material states 8.5 billion parameters for the base model.

Who publishes it?

Standard Intelligence is named as the open-source publisher.

Last reviewed: 2026-11-15

More AI audio tools tools

View all →

How we review