Neural Codec Called 'Milestone' for Digital Voice
/It's being heralded as a milestone in the long-overdue evolution of speech quality for land-mobile radio systems - the use of an adaptive neural network that replaces traditional signal processing.
A digital voice milestone was announced at the recent acoustics and speech conference in California, where the Free DV Project's David Rowe, VK5DGR, co-presented a paper describing a neural network that replaces traditional signal processing with machine learning.
“the first known real-world deployment of a neural codec – an important milestone that the Ham community can be proud of.”
Rowe and programmer Jean-Marc Valin presented the details to attendees at the IEEE Signal Processing Society conference, where David said it was well-received.
“provides unprecedented speech quality and robustness for VHF/UHF land mobile radio applications.”
Instead of using the fixed algorithms of traditional digital voice, the FreeDV Radio Encoder, known as RADE V1, employs fully adaptive machine learning to produce a higher-quality result, developed using open-source software.
The FreeDV Project - https://freedv.org/
