Hacker News Reader

Comment by crazygringo | Hacker News Reader

> 44100khz, 16-bit audio can perfectly reproduce audio as far as we can physically tell.

I agree on the kHz (as well as on MP3), but I deeply disagree on 16 bits.

Because yes, if you keep your headphone volume at a single reference level and never turn it up, then 16 bits is fine. This is very much proven.

BUT this ignores the fact that people often turn up the volume a ton to hear the quiet part of the classical music, or on that YouTube video where the volume is inexplicably 5% as loud as it should be.

So in practice, 24-bit audio allows you to retain perfect fidelity even when you have to turn the volume up. 16-bit doesn't.

I don't understand why nobody ever talks about this. (Or why you have to install special utilities on your Mac to be able to turn up the volume to 200% or 400% in order to listen to those YouTube videos that are maddeningly recorded at 5% volume.)

View on HN

You're right, it's true that 24-bit reduces the noise floor and extends the dynamic range available. However, 16-bit audio already has a range of -96db (for reference, a quiet recording studio typically has an ambient noise floor of around -60db). In practice, this is beyond the noise floor of even the very best hi-fi systems. As you turn the volume up, you will start hearing the noise floor of your equipment long before you hear the noise floor of 16-bit audio.

Unless you mean that 24-bit allows for representing audio that is stored at an extremely quiet level at the peaks, wasting most of the dynamic range. That would make more sense - but if audio is printed in such a flawed way, I would expect other quality issues to be present as well.

"the effective dynamic range of 16 bit audio reaches 120dB in practice" https://people.xiph.org/~xiphmont/demo/neil-young.html#toc_1...

That analysis misses that when you dither you sacrifice effective sampling frequency for dynamic range. 44.1KHz/16bit can represent that dynamic range, but it can't represent that dynamic range at a 44.1KHz sample rate.

It doesn’t need to. The ~80dB of dynamic range that a human ear can theoretically heard is at fairly low frequency of ~2-4kHz. Dynamic range drops off considerably at higher frequencies.

In fact, the upper limit of ~16kHz is defined by the intersection of the “threshold of pain” power curve and the “threshold of hearing” curve. So the human ear has zero dB of dynamic range at the upper frequency limit.

Ok, but what's the shape of those curves? I could believe that you can dither to adequate dynamic range and still have a high enough sampling frequency across the entire frequency range, but you'd have to actually do that calculation and show it. Also we don't just listen to pure tones - if I have a passage that includes both 12kHz frequencies and 4kHz frequencies with a bunch of dynamic range, are you going to be able to dither that without losing the high part?

Why does dithering sacrifice "effective sampling frequency"? You're just adding extremely small amounts of white noise to reduce quantization distortion or (in more advanced cases) noise with a power spectrum that puts the power of the noise mainly in parts of the audio spectrum that humans hear poorly.

This is not that (or it's an extreme special case of that); I'm talking about the thing the grandparent link is suggesting, representing amplitudes below 1 by having the sample sometimes be 0 and sometimes be 1. If you represent a waveform that should be 0.5 0.5 0.5 0.5 0 0 0 0 -0.5 -0.5 -0.5 -0.5 0 0 0 0 by doing 1 0 1 0 0 0 0 0 -1 0 -1 0 0 0 0 0, then yes you've increased your effective bit depth by 1, but you've halved your effective sample frequency.

What I described is what dithering always means in the context of audio applications.

I haven’t done the math, but I wouldn’t be utterly shocked if undithered 16-bit audio, cranked up some silly amount (such that full scale is 130dBA perhaps) has an audible noise floor.

This is consistent with my other comment about badly encoded MP3 being far from transparent.

It takes a certain amount of analog circuitry to bring a signal up to 130dBA full scale.

Doing the math on the analog bits, the engineering data indicates the analog noise will always end up greater than the 16-bit floor.

"One day" I built an analog preamp which had lower noise than a CD could reproduce anyway.

We’re talking about bad audio, though — setting full scale to 130dB doesn’t mean that the system can reproduce 130dB accurately or even hit 130dB at all.

In any case, while 130dB is a bit excessive, a highly efficient speaker (e.g. the old Voice Of The Theater) connected to a good modern amplifier and DAC could easily be cranked to 118dB with a noise floor that’s, at least in principle, inaudible in any normal room.

(I’m not saying this is a good idea. But seriously, check out the performance of the top amplifiers at audiosciencereview.com — these things have ridiculous performance and aren’t even that expensive. About 120dB SNR at over 100W is something you can just buy, for about $1500.)

(I’m also not claiming anything about linearity of the system or of people’s ears. But I can imagine 16 bits being put to better use in a well-considered floating point system than as plain linear PCM.)

Yeah, 18-20 bits make sense in loudly tuned cinemas.

There's nothing inherently weak about the fidelity of 16-bit audio on its own. PC audio subsystems don't deliver the full dynamic range on a single audio channel by default. They reserve headroom so that they can mix additional audio sources with less risk of clipping. Audio players that let you increase the volume beyond 100% are just letting you use the full range.

None of this is relevant to a real, dedicated music playback system that doesn't contain a digital mixer. You can't hear noise at -96dB. Your amplifier will swamp that with it's own internal noise sources. In the 80s the audiophools loved to complain that CDs were too quiet because their beloved LP noise was supposed to be desirable for some whack reason.

Most professional (i.e. cinema or public address) amplifiers have an SNR of around 108db. A properly functioning amplifier is never a likely source of noise.

People talk about dynamic range compression all the time.

Reply on news.ycombinator.com