>>101651593
yeah something like that lol
>>101651606
on my server. metadata goes into mongodb. it's non-commercial research, the laws here allow it. the audio is encrypted which could be considered DRM and circumventing DRM is illegal, however, because the audio is decrypted as standard i'm technically not circumventing anything, i'm making the same request for the decryption key that happens on playback and simply saving the audio instead of playing it. also some files are not actually encrypted
the worst that could happen is a dmca takedown notice if were to release the audio publicly, even then it would be down to each record label/publisher, and receiving a dmca isn't exactly an issue unless you refuse to take it down then the complainant can start court proceedings. i don't particularly plan on public release of the audio, i'd share it directly with other researchers that want it
i plan on training a llm for lyrics and i will allow it to work with artist names, imo there's no copyright issue there, any human can freely learn to write in the style of taylor swift for example, plus there are exceptions in copyright law for parody and so on
also if you think about it, spotify would love ai generated music to take off so they can pay even less to artists than they already do
i understand your concerns but i've probably thought about it more than you, i've literally had dreams where i'm discussing ai training and copyright
>best possible audio format
here's the distinct values for file.format
['AAC_24',
'MP3_96',
'MP4_128',
'MP4_128_DUAL',
'MP4_256',
'MP4_256_DUAL',
'OGG_VORBIS_160',
'OGG_VORBIS_320',
'OGG_VORBIS_96']
the best is either ogg 320 or mp4 256. mp3 96 is the short preview
spotify hi-fi still isn't out
