Mathematically lossless codecs are quite boring actually. The best and the worse are pretty close to each other. They don't have any kind of psychoacoustic model in them which is another reason they are boring. They also become less efficient the nosier the content even though we don't care much about the noise.
Prof, please correct me if I'm wrong here, I'm trying to understand this in simplest terms for my own sake:
- mathematical lossless-ness appoach attempts to preserve all information that was recorded regardless of whether it has any affect on the listener, even the noise
- perceptual lossless-ness attempts to preserve only what is needed to make the information indistinguishable (for humans) from mathematical lossless-ness (minus noise?)
- this is done by applying psychoacoustic model to the information and for the purpose of reducing the amount of data that needs to be stored/transmitted
If my understanding is correct, while interesting, MQA doesn't seem to solve any problem in our 4k streaming/cloud storage age. (Yes, I'm slow.. thank for being patient with us smooth brained folks
) Would it not be more effective to address noise issues at the recording and mastering level? Rather than applying a universal model to all different noise situations? Or do we have (/will soon have) adequate models for distinguishing between what is noise and what is music?
Here's a really "out-there", follow-up thought: Is the endeavor of psychoacoustic modeling even relevant when we are steadily marching towards readily available human modifications? e.g. a cochlear implant that extends your frequency perception - Would our mushy processors even be able to compute that? Or will it completely ruin our perception of music? From what I've seen of Amir's videos on high-res files and their ultrasonic information, my intuition tells me that as long as the ultrasonics behave in the same predictable way as music in the hearing range our brains should enjoy it. monkey brain likey patterns
I naturally wonder if Amir could use his 'tricks' (not really tricks, just very not-normal listening regimes) and training to detect differences between MQA encoded and PCM source. For example, for 16 bits vs 24, a 'trick' is to select an extremely quiet part of the music, and listen to it at a very high playback level. A level that would be deafening for most of the track. To detect 320kbps (as well as, I presume, very high VBR) mp3 encoding by a high quality encoder , a trick is to use a 'killer' musical selection, difficult to encode, and/or to zero in on a micro-instant that 'tells' and play that over and over until you learn to hear the difference.
I wonder how MQA would fare in such tests that employ very non normal listening....and would it still qualify as 'perceptually lossless' if it failed them?
yes, please!
Also, I'm really not enjoying the engineer/scientist/economist bashing subplot going on here... Just completely irrelevant and not at all in the spirit of this community. Lots of random hot takes on people's identities rather than the substance of their work. We are here to come to a
scientifically grounded consensus on the merits of MQA.
If you are not an expert in this field and are jumping to conclusions based on who the researchers are, please refrain so the people who have the technical knowledge can take the time to apply due diligence in their scrutiny.
We need to be patient, y'all. Look, I've had my MQA pitchfork sharpened long before this food fight - same as most here... ASR is not built so we can sit on top and stroke our egos, congratulating each other for being right. This is the largest well of audio science knowledge on the anglophone web and Professor M didn't build its reputation on jumping to conclusions.