[FR] Preprocess voice in the client and only send data when spoken

[HarikalarKutusu]

Currently, all sounds are streamed continuously, which also hogs the server. In a normal chess play there are 30-50 moves * 2 per game in 20-120 minutes. Assuming 30 minutes/game with 40 moves/player and 4 secs per command, that would mean less than 10%.

If we can implement this:

• Any silence, background noise etc will be filtered out.
• Only relevant data will be sent, so less communication on both sides
• We can use the same server for multiple connections by queuing incoming packets. In this case, a server can become a "language server" with (say) 5-10 connection/inference capacity (which can be adaptive). This is as indicated in [FR] Increase server resources - the scaling problem #4 , third option.

This can be achieved by measuring the sound level/energy with pre-buffering.

One downside to this is: The client will be processing audio continuously, which can be bad for future mobile users in terms of battery usage.