Standard Intelligence has released an open-source audio-only transformer base model called hertz-dev, with an impressive 8.5 billion parameters. One component, hertz-codec, outperforms other codecs at a lower bitrate and has a unique architecture. Hertz-dev also includes a transformer decoder and a transformer stack with unparalleled low latency. This groundbreaking model is a glimpse into the future of real-time voice interaction and is designed to be easily fine-tuned for various tasks. The potential for Hertz-dev to interact in a human-like manner sets it apart from other models, making it a valuable tool for researchers.
https://si.inc/hertz-dev/