Something a lot of people knew was coming for ages, but finally seems to be here as a product.
Compression through machine learning models for voice calls.
Lot of previous work in creating models that do voice transfer, so this was coming eventually.
I think it’s misleading to call this compression, it’s transcribed audio, the medium is changed entirely (and fwiw in an incredibly lossy way, losing tone, cadence and style). Opus 1.5 uses ML and compression to effectively transfer speech audio data at 48kb/s with even 80% packet loss. That’s something I would think impacts voice calls - I don’t understand how this does any more than a browser “Live Chat” application, if you could enlighten me?
Something a lot of people knew was coming for ages, but finally seems to be here as a product. Compression through machine learning models for voice calls.
Lot of previous work in creating models that do voice transfer, so this was coming eventually.
I think it’s misleading to call this compression, it’s transcribed audio, the medium is changed entirely (and fwiw in an incredibly lossy way, losing tone, cadence and style). Opus 1.5 uses ML and compression to effectively transfer speech audio data at 48kb/s with even 80% packet loss. That’s something I would think impacts voice calls - I don’t understand how this does any more than a browser “Live Chat” application, if you could enlighten me?