Add more info to CPU warning logs from the Voice Focus worklet

[hanfangAmazon]

The log CPU warning: cpu doesn't provide much information under the hood, and the warning is triggered on every 10 audio chunks skipped which isn't ideal.

Consider checking percentage audio chunks skipped per second, and if it's greater than a threshold then log the message, example

Voice Focus CPU warning: skipped 5.2% audio chunks

[richhx]

Hi hanfengAmazon@, thanks for bringing this up. We can look into adding this to our backlog. We do send the number of chunks that are skipped (i.e. 10 as you mentioned) in our worker payload and we can log that. I think getting a percentage is tricky because we only process the latest buffer at a time and if CPU is falling far behind, the percentage will be based on only what we know so far. I'd prefer to not have it loop over all buffers as it'll take up more bandwidth.

Is there a particular reason you would like a percentage logged? I agree that the current log message isn't quite useful, and I can look into updating it to log the number of chunks missed. I think the most useful thing is that the worker is not able to keep up in processing, and we offer a delegate.onCPUWarning callback, which can be used to just be notified that Voice Focus is not able to keep up.

[hanfangAmazon]

@richhx Thanks

Is the number of chunks VF processed per buffer always the same? Assume that we have 1,000 chunks in the latest buffer, and only 10 chunks are skipped, which isn't a big deal. In case of only 100 chunks in the buffer, 10 skipped chunks has greater impact.

That's why we want a percentage, it doesn't necessarily to be percentage skipped per second, it can also be percentage skipped in latest buffer, or any other stats data that's not based on a fixed value.

[richhx]

The number of chunks processed per buffer is usually finite for a given render quanta, but it can accumulate if the worker is falling behind, so makes sense to potentially add some way to measure it better. I've added it as an action item in our backlog to look into further.