3 ms·
I wonder how big the codec itself is. Neural networks are not exactly small. Also, the processing latency - Lyra mentions on its GitHub page 100ms.
by userbinator 5y ago
I wonder how big the codec itself is. Neural networks are not exactly small. Also, the processing latency - Lyra mentions on its GitHub page 100ms.
- sanxiyn 5y agoThe paper says 8.4 million parameters for the default model. Assuming 4 bytes per parameter, that's about 30 MB.
- Scoundreller 5y agoBeing able to download in advance is huge. Looking forward to satellite audio for everything spoken-voice one day. Some org was working on it, but they were working with 2kbps 24/7, which is definitely not a lot!
- benob 5y agoIn the papier they compare lattencies from 7ms to 26ms and observe no loss of quality. Processing requirement is lower for higher latencies because of batching.