AI Signal 415
Meta EnCodec compresses 2.9MB song to 21KB for paper QR storage and LoRa transfer
↗
A maker used Meta's EnCodec neural audio codec to compress a 2.9MB MP3 to roughly 21KB, small enough to print as eight QR codes on paper or transmit over LoRa radio between ESP32 boards.
Neural audio codecs achieve compression ratios far beyond conventional methods, enabling storage and transmission media previously impractical for audio. However, the compressed tokens require a specific neural decoder for playback, creating a preservation dependency where the physical medium may outlast the ability to decode it.
Written by elseif from the cluster below · every claim links back to a sourceThe three things worth knowing
Meta's EnCodec compressed a 2.9MB MP3 to roughly 21KB at its 3 kbps setting, preserving acceptable audio quality, while the 1.5 kbps setting produced heavily degraded output.
The 21KB file fits across eight QR codes printed on paper with no error correction redundancy, or transmits over LoRa radio in roughly 90 minutes using a deliberately conservative ACK-and-retry protocol for duty-cycle compliance.
Playback requires EnCodec's exact 24 kHz decoder, meaning the stored tokens become unreadable if that specific neural network becomes unavailable, echoing the obsolescence risk of the 1985 Cauzin Softstrip format.
THE CLUSTER