What is the minimum bit-depth to get good audio quality?
Note: I replaced the song with the WAV version
Found here: https://www.head-fi.org/threads/10-bits-...lf.731964/
Some other interesting links:
...and some new ones:
According to this paper
http://decoy.iki.fi/dsound/ambisonic/mot...oding2.pdf
to achieve audible transparency, PCM with following features are needed: - 58kHz and 14 bit with noise shaping
OR
- 48kHz and 20 bit without noise shaping
in this article
https://sonicscoop.com/2013/08/29/why-al...bly-wrong/
some analog sources are listed with dynamic range and related bit depth "resolutions": - 78 RPM: 30-40 dB ➔ 5-6 bits
- cassette tapes (prerecorded): 40 dB ➔ 6 bits
- cassette tapes (recorded, best case scenario): 70dB ➔ 11-12 bits
- vinyl: 60-70dB ➔ about 11 bits
I can add to them: - laserdisc (analog CX off): 50-58 dB ➔ 9-10 bits
- laserdisc (analog CX on): 62-74 dB ➔ 10-12 bits
- VHS (mono non HiFi): 45 dB ➔ 6 bits
- VHS (stereo HiFi): 80 dB ➔ 13 bits
few considerations after some tests I made: - at 16 bits I can hear -94 dB sine wave at reference level (and below) on my amplifier, possibly because of its sound nature (read: annoying) while I can barely hear classical music at -70dB
- 11 bits (decimated from 16 bits) are enough to retain very good quality with a low (if not not audible) noise floor
- 8 bits (with or without noise shaped) has too much noise for my taste
- I just remembered that Akai 80s samplers used 12 bits sampling rate (and they had good reasons)
My conclusion: 16 bits has around 96 dB dynamic range, that could top 120 dB and more using noise shaping; they are more than enough to cover all music and movie dynamic range when listening even to reference level (105 dB), on a quiet room (or even recording studio, I guess); about frequency, considering that some very special gifted person that could hear frequencies higher than 24kHz could exist, but they must be extremely rare - more or less like tetrachromatic women IMHO
I'd say that 16bit/48kHz for release is good enough for 99.9% of situations/persons; productions wise, I'd stay with 24bit to take in account multiple processing and let the noise floor stay as low as possible, while 60kHz would be ideal to reproduce ANY frequency that human beings could hear; as industry standard is 96kHz, using it would be not harmful, apart using twice the space.
So, 24bit/48kHz for production is again enough for 99.9% of situations/persons IMHO, but you can go up to 24bit/96kHz to cover the residual 0.1%; anything more is overkill.
I think the consensus is that working in 24bit and mastering in 16bit is 'good enough' for most purposes. I know that the LFE channel for AC-3 and such was designed amongst other reasons to increase the effective dynamic range to 20bit. I can't imagine anything ever utilising the full dynamic range of 24bit (safely anyway)
[links moved to first post]
I'm the only person I know (offline anyway) who actually owns SACDs so I've done a fair bit of reading about DSD, it's kind of a s***show. The whole thing is completely reliant on noise shaping.
I've read that Ken Rockwell page before but I read it again because I find his stuff as entertainment as it is insightful. His bit about audiophiles is highly amusing...
Moved all links to the first post, and added some new ones.
Worth noting that all of these considerations are for linear PCM. You could theoretically apply a curve that is then decoded upon playback. There are actually PCM types that do this, but they're meant for nothing more than retaining voice intelligibility, so the quality is too low for practical purposes. But I imagine you could gain a bit of dynamic range safely with a "gamma curve". Also ... splitting into multiple bands and then doing that? 
Anyway, my 2 cents ... I'd go 24 bit always, even for publishing. Noise shaping/dithering and all that is kind of a lie imo. With good headphones you can clearly hear the added dithering noise, whether shaped or not. In fact I once did a test and I liked the shaped noise less because it was kind of high and annoying where the normal one was more equal across the frequencies ... but I will acknowledge that this is more relevant for headphones probably than for real world situations. But coming from that, I've retained some skepticism about 16 bit.
Also this all depends on your reference level. Aka just how loud is the highest peak?
The only times I'd personally stay at 16 bit is if the original audio was 16 bit to begin with, like with digital LPCM sources like a Laserdisc at 16 bit. Though ... sometimes I wonder if the resampling alone might not introduce some kind of noise that adds to the noise floor. But I haven't done a test about this.
https://wiki.xiph.org/Videos/Digital_Show_and_Tell
Watch the video, is an eye opener!
Waiting for your comment after you watched it... let's say, in about half an hour?
Not in the mood to watch that, but it's possible I've seen it before.
OK, but I strongly advice to watch it ASAP - you know, I was a "believer" once, but now, with quite some experience, I follow only one rule: "TRUSTNO1"! That means, do not trust me, nor the authors of this video or of the other posts and papers I linked, but trust not also the marketing... the latter at least has something to earn to lie!
|
