Neural Audio Compression Enables 2.9MB Song to be Printed on Paper as Eight QR Codes
A maker has successfully compressed a 2.9MB song using Meta's AI codec and printed it on paper as eight QR codes, requiring a neural network for playback.
## Neural Audio Compression Breakthrough A recent innovation in neural audio compression has enabled a maker to compress a 2.9MB song by 1000 times, allowing it to be printed on paper as eight QR codes. This achievement was made possible by Meta's EnCodec neural audio codec, which was used to convert the waveform of the song into discrete tokens that can be decoded back into audio. ## The Compression Process The compression process involved using Meta's EnCodec to convert the 2.9MB MP3 song into a 21KB file of latent tokens. This reduction of 1000 times was achieved by using a neural network to analyze and compress the audio data. The compressed file was then split across eight QR codes, each carrying a 1-byte ID and a chunk of the token stream. ## Printing and Playback The eight QR codes were arranged as a double-sided "paper cassette" and printed on paper. However, it's worth noting that the song can only be played back using a neural network that can decode the EnCodec tokens. This means that the song will only survive as long as the neural network that can decode it. ## Wireless Transfer The compressed file was also beamed between two ESP32 boards over LoRa radio, taking around 90 minutes per song. This was achieved by using an ACK-and-retry protocol to ensure reliable transfer of the data. ## Historical Context The idea of printing software on paper dates back to the Cauzin Softstrip of 1985, a 2D barcode that allowed magazines to publish programs at up to 5,500 bytes per strip. However, this technology faded once floppy disks became widely available. The current innovation builds on this idea, but with the added complexity of requiring a neural network for playback. ## Conclusion The successful compression of a 2.9MB song using Meta's EnCodec neural audio codec and the subsequent printing of the compressed data on paper as eight QR codes is a significant achievement. This innovation has the potential to enable new forms of audio storage and playback, and it will be interesting to see how it develops in the future.
