Learn more about our continuing work and take a look at the playable demo in this informative article .

This moment, we turn LPCNet into a really low-bitrate neural speech codec which ’s really usable on present hardware and even on phones (as clarified within this paper). It’s the first time that a neural vocoder is able to operate in real-time using just one CPU core on a phone (instead of a high-end GPU)! The resulting bitrate — just 1.6 kb/s — is roughly 10 times less than what wideband codecs typically use. The quality is significantly better than existing quite low bitrate vocoders. In fact, it s akin to that of more codecs using high bitrates.


This new codec may be used to enhance voice quality in countries with poor network connectivity. In addition, it can be utilized as redundancy to improve robustness. In storage programs, it may compress an hour-long podcast to just 720 kB (so that you ’ll still have space left on your floppy disc ). With some additional work, codecs can help enhance at low bitrates.

This is an upgrade on the LPCNet project, an efficient neural speech synthesizer in Mozilla’s Emerging Technologies group.
Screenshot of a demo player that shows the caliber of LPCNet-coded address