Martin Evans

Perhaps try putting some logging into the [Audio Decoding Buffer](https://github.com/martindevans/MumbleSharp/blob/master/MumbleSharp/Audio/AudioDecodingBuffer.cs) to see if it's running out of samples to play back.

Yeah basically encoded audio gets dumped into this buffer when it arrives from the network and then the audio devices pulls from it (in `Read`) when it needs something to...

That's definitely the problem then, every time `FillBuffer()` returns false is a failure that results in audio artifacts. Fixing it ideally requires a jitter buffer (as I described [here](https://github.com/martindevans/MumbleSharp/issues/46#issuecomment-513055783)). A...

I'm afraid I can't remember the details of the playback pipeline (it's been a long time since I wrote this). I'd suggest finding all uses of AudioDecodingBuffer and tracing it...

Ah, that's a bigger issue than a simple jitter/timing problem! Which application are you using to play back audio (the core MumbleSharp library itself doesn't handle playback iirc)?

1. Ideally rather than using real time (stopwatch or DateTime) you'd be using the sample counts as that's the measure of time the audio system cares about. It also makes...

Sure, it sounds like it's at least an improvement on what we've got!

> It's a fine proof of concept or starting point, but it's a development project not something to take and use. That's a very fair assessment of the state of...

That's a good point. The purpose of making those internal was so that people using the models can't fiddle with the values, but of course you're right that makes implementing...

I've been investigating the performance of models with batched inference. I had expected slightly different results based on the number of parallel sequences being evaluated (i.e. some small amount of...