5 ms·
Next step: train a NN to predict what we’re going to say to compensate network latency :)
by ngrilly 5y ago
Next step: train a NN to predict what we’re going to say to compensate network latency :)
- vletal 5y agoHave you ever played a counterstrike with a lagging network? Observe the teammates running into walls and such. That's what the conversation would sound like.
- meowface 5y agoIt could make for a hilarious new comedic art form.
- vbezhenar 5y agoRunning into a wall is not a result of NN prediction. It's a result of naïve linear prediction. NN will predict that he'll stop, turn around, may be shot someone. The hard thing is to not disappoint NN by actually running into a wall.
- vletal 5y agoYou need to take into consideration the complexity. Counterstrike has very tiny set of possible actions you can take. Compare that to the space of things one can say. A NN is going to fail on that at least as missereably as linear interpolation fails in CS.
- hypertele-Xii 5y agoWhen speech prediction fails, it should sound clearly wrong. Otherwise we risk serious misunderstanding when the prediction says something that sounds good but means different. We already have trouble like this with texting autocorrection.
- lxgr 5y agoDon‘t modern codecs already do something like this to conceal packet loss? (Obviously only at the phoneme and not word or semantic level.)
- amelius 5y agoDoes this work for the starting phoneme of a word?
- pbhjpbhj 5y agoJust running with that a little ... So, at what point during a phoneme does it become distinct? After any one phoneme has been sent down the wire you can truncate the audio and send a phoneme identifier; at the other end they just replay the phoneme from a cache. Like doing speech to text, but using phonemes, and sending the phoneme hash string instead of the actual audio. Must have been done already?
- ngrilly 5y agoWondering the same.
- mchusma 5y agoGoogles duo already does this. In my experience, it's amazing. http://ai.googleblog.com/2020/04/improving-audio-quality-in-duo-with.html http://ai.googleblog.com/2020/04/improving-audio-quality-in-...
- ngrilly 5y agoI remember reading that paper. This is amazing. But it's not really predicting what we're gone say beyond the current syllable: > While our model learns how to plausibly continue speech, this is only true on a short scale — it can finish a syllable but does not predict words, per se.