IPB

Welcome Guest ( Log In | Register )

 
Reply to this topicStart new topic
Dolby B simulation
audioapprentice
post Nov 12 2008, 03:24
Post #1





Group: Members
Posts: 58
Joined: 19-August 08
Member No.: 57250



I've always wanted a software Dolby B encoder/decoder for:
(a) recreating a period feel and
(b) for processing tapes digitized with Dolby turned off.


After having read the whitepapers from the Dolby website here
Dolby B Technical Details, I produced a single band 2:1 bilinear compressor (for the encoding side) and matching expander (for the decoding side) in CoolEdit:
Unity gain above -20dB
2:1 Compression from -20 down to -40
Fixed gain (10dB) below -40

However, the sliding band filter has me confused. Sticking to the encoding side for now:
Would multiple single band compressors be the same as a sliding band compressor? For example, a series of compressors that have bands of say 1kHz, 2kHz, 3kHz...20kHz respectively.

Any comment, corrections, advice much appreciated.
Go to the top of the page
+Quote Post
Slipstreem
post Nov 12 2008, 04:14
Post #2





Group: Members
Posts: 966
Joined: 7-July 06
Member No.: 32660



I can't directly answer your question as I don't have filter knowledge within the digital domain, but I think you're ideally looking for a fixed frequency filter with a gentle slope of dynamically adjustable gain. You're only trying to simulate a few simple C-R networks and a stereo compander or two, not banks of 12th-order Chebyshev's. tongue.gif

This is a very interesting topic for me also as the only thing that bothered me about my latest cassette to MP3 session with Audacity was that it revealed that the Dolby NR systems on the different recording and playback decks were not calibrated quite the same. Sometimes an excess of HF cut was applied on playback and sometimes not enough. I could clearly hear the music "breathing" at times on some content.

The incredible truth revealed itself when looking at spectrograms of the interim WAV files in 32-bit float at 44.1kHz. There was still significant stable HF content to be retrieved from these tapes all the way up to 20kHz after 15+ years of OK-ish storage.

A tweakable DSP to do this with Foobar2000 would be absolutely brilliant and a massive help to a lot of users, especially if it had profiles for DolbyB, DolbyC, and even DolbySR. A person could then Dolby encode source material prior to transfer onto tape and then de-emphasise it correctly on playback.

Have you ever heard how good a decent cassette deck could sound if it had DolbySR? Neither have I. I'd like to hear it though. I can test it on my 3-Head TEAC if you like. biggrin.gif

Cheers, Slipstreem. cool.gif

EDIT: You may want to emulate the Rocktron HUSH2000 NR system too. A full datasheet is only a quick Google away. It shouldn't be any harder to implement than DolbyB by the looks of it and offers single-ended dynamic NR of up to 25dB (adjustable) on analogue source material whilst maintaining reasonable to good perceptual transparency. I struggle to ABX it at up to 10dB of NR on my best taped source material, and even that's well worth having.

Don't forget that HUSH2000 implements many of the basics of an analogue psychoacoustic model in terms of perceived noise floor calculation, perceived average loudness and dynamic filtering bandwidth requirements in real-time with, arguably, infinite resolution. It tracks a lowpass filter with available content in the input spectrum with carefully calculated attack and decay times to make it as transparent as possible in operation. The signal then gets a 1:2 dynamic expansion with tightly controlled attack and decay timings below a preset level that references the actual source audio on-the-fly to determine the best auto-threshold in relation to the actual noise floor.

That would make a handy tool in itself, especially for those lucky enough to be remixing old multi-track masters on a budget, or just wanting to clean up tapes not pre-emphasised for any form of NR on playback but doing it as discretely as possible. It even de-noises vinyl to an extent and I seem to remember a few of the time constants being related to ripple rejection to remove the effects of rumble and record warp on the detector stages.

One thing concerns me though and this is due to personal ignorance rather than knowledge in my case, but would a typical modern CPU have the horsepower to do all of this in real-time (preferably all in 32-bit float internally if that's possible) with sufficiently fine granularity along the time axis to perfectly match or beat the performance of the physical HUSH IC? They were never particularly expensive when available, but they did help to resolve a simple problem simply but elegantly and would be well worth chucking a few horsepower at if a person needed to do it "properly" as a one-off when transferring tapes to digital.

You can use it in conjunction with Dolby encoded recordings AFTER they've been natively de-emphasized, so it's possible to get the perceived SNR of Metal cassette tapes close to 20dB nearer the noise floor with Dolby B daisy-chained with 10dB of HUSH than with no NR at all. Couple native DolbyC with 10dB of HUSH and you're up to a perceived noise floor at around -85 to -90dB.

Just my mad ramblings... smile.gif

This post has been edited by Slipstreem: Nov 12 2008, 05:39
Go to the top of the page
+Quote Post
Glenn Gundlach
post Nov 12 2008, 04:39
Post #3





Group: Members
Posts: 359
Joined: 19-April 08
From: LA
Member No.: 52914



QUOTE (Slipstreem @ Nov 11 2008, 19:14) *
I can't directly answer your question as I don't have filter knowledge within the digital domain, but I think you're ideally looking for a fixed frequency filter with a gentle slope of dynamically adjustable gain. You're only trying to simulate a few simple C-R networks and a stereo compander or two, not banks of 12th-order Chebyshev's. tongue.gif

This is a very interesting topic for me also as the only thing that bothered me about my latest cassette to MP3 session with Audacity was that it revealed that the Dolby NR systems on the different recording and playback decks were not calibrated quite the same. Sometimes an excess of HF cut was applied on playback and sometimes not enough. I could clearly hear the music "breathing" at times on some content.

The incredible truth revealed itself when looking at spectrograms of the interim WAV files in 32-bit float at 44.1kHz. There was still significant stable HF content to be retrieved from these tapes all the way up to 20kHz after 15+ years of OK-ish storage.

A tweakable DSP to do this with Foobar2000 would be absolutely brilliant and a massive help to a lot of users, especially if it had profiles for DolbyB, DolbyC, and even DolbySR. A person could then Dolby encode source material prior to transfer onto tape and then de-emphasise it correctly on playback.

Have you ever heard how good a decent cassette deck could sound if it had DolbySR? Neither have I. I'd like to hear it though. I can test it on my 3-Head TEAC if you like. biggrin.gif

Cheers, Slipstreem. cool.gif


It's a little worse that you think. Dolby B is the top processor from Dolby A. It has a sliding turnover frequency, IIRC 500 to 2KHz and its a compander in that low level HF was boosted up to 10 dB and loud HF boosted 0 during record encoding and the reverse during playback. Your processing has to change both the gain and the turnover frequency to properly decode it.

The JVC ANRS system from the early '70s used a fixed frequency turnover and changed only the gain. IIRC the turnover was at 1KHz but I don't know if the gain spread was 10 dB or something else. It was touted as being 'compatible' with Dolby B.

The processing I really disliked was Dolby C. I believe the turnover was at 200 Hz and the gain spread was 15 dB. I don't know if the frequency was static or moving with level.

I'd like to Dolby B decode in software rather than hardware so that I can apply group delay processing to correct the squarewave response on OLD 1/4 " open reel tapes. Since the group delay happened in the tape deck after Dolby encoding, I'd like to phase correct _then_ Dolby decode but my Dolby B processing is an external analog processor (Advent 100A or Sony NR-335) and I'd like to avoid the multiple A-D / D-A conversion that would require. Maybe I should just try phase correction after Dolby decoding.

G
Go to the top of the page
+Quote Post
Slipstreem
post Nov 12 2008, 06:29
Post #4





Group: Members
Posts: 966
Joined: 7-July 06
Member No.: 32660



QUOTE (Glenn Gundlach @ Nov 12 2008, 03:39) *
It's a little worse that you think. Dolby B is the top processor from Dolby A. It has a sliding turnover frequency, IIRC 500 to 2KHz and its a compander in that low level HF was boosted up to 10 dB and loud HF boosted 0 during record encoding and the reverse during playback. Your processing has to change both the gain and the turnover frequency to properly decode it.
I was forgetting that. So Dolby B and the single-ended HUSH2000 were more similar than I thought and it's all about attack and decay times for companders and accurately tracked rate of change on the adjustable lowpass filter. One is "programmed" to work optimally as an encode/decode process and the other is mild enough to be largely unobtrusive but still manages around 10dB in perceived SNR improvement running single-ended for the cusp of perceptual transparency at that dynamically chosen bitrate. That one would be fun to play around with in software in real-time.

QUOTE
The JVC ANRS system from the early '70s used a fixed frequency turnover and changed only the gain. IIRC the turnover was at 1KHz but I don't know if the gain spread was 10 dB or something else. It was touted as being 'compatible' with Dolby B.
And how I loved it over my early genuine DolbyB equipped decks at the time. I had a few based around the Dolby-licensed National Semiconductors LM1112(?) IC and sibilants were all torn to shreds on many an early DolbyB deck using these Dolby-approved ICs during the recording phase, especially if you stuck a nice crispy CD source into one. It hurt their reputation quite badly in my eyes and made me a little sceptical of Dolby as a competent audio processing company for a while, TBH. The constant lisping managed to make even Barry White sound camp at times. You couldn't fix it once the recording was made, but I did come up with a few cheap component changes around the ICs to fix it at a later date so no more broken recordings were made after that. ANRS never suffered from that and worked better for my ears anyway. I got on fine with all of the DolbyB incarnations after that on other decks with a fine bias trim control, either factory-fitted as standard or modified/fitted by myself, even if it meant better recordings for almost no zero additional outlay and an extra twiddly knob concealed around the back of the tape deck to keep up appearances if nothing else. tongue.gif

QUOTE
The processing I really disliked was Dolby C. I believe the turnover was at 200 Hz and the gain spread was 15 dB. I don't know if the frequency was static or moving with level.
I've only had experience of that working entirely to my liking on one particular deck (although I guess there are plenty of others) and that's my current 3-Head TEAC with Chrome or Metal tapes where the additional help of DolbyHX Pro during the recording phase helps to linearise the non-linearity of the tape to a large degree. DolbyC really does need DolbyHX Pro along for the ride to work as transparently as possible on cassette tape recordings at a tape speed of only 47.6 mm/second, and not all early DolbyC equipped decks had DolbyHX Pro as well. DolbyC without DolbyHX Pro when recording = fail! Any recordings made with both will play back identically whether the playback deck is equipped with DolbyHX Pro or not. That's a single-ended process at recording time and there is no de-emphasis required on playback. In other words, take magnetic tape out of the equation and I think that DolbyC is probably quite transparent and could possibly improve the perceived noise floor by as much as 25dB.

Wouldn't it be great to have a tool that could replicate previously established analogue-invoked algorithms to perfection for use with perfect or imperfect digital media? Imagine actually hearing DolbyC, DolbySR, JVC ANRS or HUSH2000 working almost exactly as intended in the laboratory with lossy digital media instead of noisy tapes. They all work very well in their own way for attenuating unwanted content in the analogue domain, so couldn't the apparent dynamic range of any simple digital audio encoder be perceptually increased by emphasising on encoding and then de-emphasising on playback to take advantage of this by giving a perceived increase in apparent SNR and maybe even perceived quality at times of near perceived silence? Would this result in a smaller lossy file in VBR for the same perceived target quality? Would it reduce audibly perceived quantisation errors at low volume levels by dynamically "weighting" the bits to better suit the audio content on-the-fly when both pre-emphasised and de-emphasised? Can a flag be set in a header to invoke a de-emphasis on playback with, say, MP3 to suit a custom pre-emphasis without breaking anything? Do we then have a new 'LAME-Super-VBR' mode making a more intuitive guess as to where the bits need spending the most for best detail and maximum possible perceived quality? Ahem! Don't feel obliged to answer that last one. Maybe it could even toggle between existing standard profiles on-the-fly if necessary to achieve the highest possible perceived dynamic range within standard profiles?

Maybe expecting it to help an 8-bit sound file much in reality is a bit optimistic in my view, but it may work well enough to make 12-bit audio sound more like 16-bit audio, and 16-bit audio sound more like 24-bit audio, perceptually.

Is this in any way helpful in conjunction with a lossy encoder?

Would it effectively expand the volume of MP3 artifacts downwards in relation to wanted source material but leave everything else in the MP3 almost completely unchanged sonically by the process?

What would a plain 2:1 compression in dynamic range during encoding accompanied by a 1:2 expansion of dynamic range on decoding do in terms of artifact level versus original signal, if anything? Would the playback end "sound" as though it had more bits of resolution available to it where needed if the MP3 is decoded and played back at 24-bit resolution now?

Bear in mind that it should be possible to produce an accurate anti-phase replica of the difference between expected signal and actual signal very quickly so it may be useful for on-the-fly automatic de-glitching if the file is sufficiently padded to give us some buffer time to generate an accurate error-correction signal or "anti-glitch". If the glitch has a known fingerprint then apply a de-glitching filter combination that specifically targets that fingerprint. If unsure then add the fingerprint to a library and link it to the best-fit current solution in that library until the next version of the decoder is released with optimised and/or additional de-glitchers. You could have an emergency attenuate or mute filter that could be triggered for up to, say, 1ms at a time to physically mute short disasters that can't currently be dealt with better in any other way and are likely to be less obtrusive as a tiny gap rather than as a big click. Internal redirects to custom user-defined profiles could lead to interesting artifact fingerprinting techniques and countermeasures being developed. The profile could be fully dynamic in all of its parameters and self-program on-the-fly in slightly delayed real-time if it knows what's coming a few milliseconds before playtime. Don't most of us have the computing horsepower required to do this on our home PCs already? Isn't this roughly how an intelligent scratch removal algorithm would work with damaged vinyl in the analogue domain?

It's only really the slightly noticeable rise in audible (to me) artifacts that stops me from dropping permanently to -V5 (or -V7 for audiobooks) from -V3 to be honest. If it's possible to get rid of most of them or even just the worst of them by superimposing a second psychoacoustic model and getting away with a little on-the-fly interpolated but quite accurate error detection and correction going on seamlessly on-the-fly during playback, then I'd be interested in being a guinea pig for some ABX-ing and a modified playback plug-in for Foobar2000 so I can play them back. Not that the files would be of any use in a conventional MP3 player, but they might be noticeably smaller and still sound the same to you or I within all of the other constraints of the MP3 specification. Maybe "Mount LAME Super" might be a better name as it's internally MP3 compliant but has to be mounted at least via a free plug-in to become accurately decodable, hence truly playable. Heh. wink.gif

Could this be incorporated into the LAME encoder later after rigorous testing in a new two-pass VBR encoding profile for LAME with a conventional VBR -Vn pass carried out first whilst also "looking" for and tagging artifact signatures, then custom correction being automatically applied during the second pass of the encoding process? Would this allow us to push the MP3 format as far as it can sensibly go in VBR for the time being by actively searching and patching with some degree of success whilst remaining perceptually transparent and within MP3 specs yet still creating some kind of Super-VBR in the process, or doesn't it work that way? If it worked very well but made the encoding process a lot slower, would it be worth the one-off overhead of a longer encoding time? How much space in percentage terms would it be likely to save you in reality anyway for the same level of artifact reduction as a step or two up the -Vn ladder in terms of average VBR bitrate? 30%? 10%? 2%?

Even just a pre-emptive strike on a second-pass by yanking the bitrate up to 320Kbps for a few frames might do the trick for many problem samples and cut out all or most of the complications mentioned above. Is that doable with less messing around?

I'd really like to see all of this put to the test with a DSP though. I just don't have the necessary skills to do it in software although I did build a quad-band stereo compander in reality many moons ago and that managed up to a perceived 12dB of SNR improvement transparently to me personally single-ended, so that's a potential for a 24dB improvement if used as a compressor for pre- and an expander for de-emphasis. Also, is a potential 24dB of artifact suppression worth the effort even if it does work in practice? Would it make all our low bitrate MP3s sound like perfectly "clean" encodings with just an upper cut-off frequency as determined by the DSP during the encoding phase? A bit in the header somewhere triggers your player to de-emphasise in, say, HUSH2000 and off it goes automatically and does it to a standard profile with almost bit-for-bit perfect emulation of the ideal equivalent analogue de-emphasis.

Could this be done in practice? Would there be any point in doing this in practice? Isn't this just a slightly different approach to the error-correction that a standalone CD player does all the time with far less horsepower than a typical desktop PC has nowadays? Does anybody ever complain about the "poor" error-correction on their standalone CD player because it's effectively masking errors in one way or another? Or, is it the first of April?

But the real questions are, "Have I just gone ever so slightly mad?" and/or ,"Am I actually talking complete b****cks?" tongue.gif

Cheers, Slipstreem. cool.gif

This post has been edited by Slipstreem: Nov 12 2008, 14:08
Go to the top of the page
+Quote Post
audioapprentice
post Nov 13 2008, 04:09
Post #5





Group: Members
Posts: 58
Joined: 19-August 08
Member No.: 57250



Quick summary of information on the process I have so far:

To encode:
1. Apply bilinear compressor with sliding frequency band.

To Decode:
1. Apply bilinear expander with sliding frequency band.

The sliding frequency band:
1. The low-level threshold is -40dB.
2. At 1kHz if signal < threshold (-40dB) band in default position (no sliding).
3. At 1kHz if signal > threshold (-40dB) slide up (the frequency range) by x kHz
Q. What increment does it slide by? It could be 2kHz as the 3kHz band (from Dolby A)

doesn't modulate noise and was the initial band they tried when developing Dolby B.

The compressor attack time is long ("tens of milliseconds"). It seems to be fixed but
EDIT: According to the discussion here the attack time is 20ms and the release time is 40ms.

I could have this wrong.

Additions, corrections, etc much appreciated.

EDIT: There is a paper on digitally similating a Dolby B analogue system here. Also the various patents for Dolby NR systems have a lot of information.

UPDATE: It's probably not really worth pursuing. I may come back to this thread from time to time to post updates if there are any. For my tape transfers I will do them all with Dolby B off as that sound is the lesser of the two evils and can be fixed somewhat with some judicious EQ and holds the possibility of improvement in the future with a fuller understanding of the Dolby B process.

This post has been edited by audioapprentice: Nov 14 2008, 04:04
Go to the top of the page
+Quote Post
2Bdecided
post May 5 2009, 11:26
Post #6


ReplayGain developer


Group: Developer
Posts: 4957
Joined: 5-November 01
From: Yorkshire, UK
Member No.: 409



Tape Restore Live includes a Dolby B simulation.

It was programmed empirically, but it works well.

Cheers,
David.
Go to the top of the page
+Quote Post
HelpfulDad
post Jun 5 2013, 15:47
Post #7





Group: Members
Posts: 3
Joined: 5-June 13
Member No.: 108504



Great link! Just what I was hoping for.

I'm going to try it out. I'm pondering the idea of putting a topic on here about my project. I have a number of Quad Reel to DVD-A .iso files of some really great titles that I believe are Dolby B encoded. Some I'm sure of, because it says so right on the box in the scan of the title. The music is brilliant, but it's thin with that characteristic high-end boost you get when you don't decode a cassette encoded with Dolby B that so many mistake for clarity. But the quality of these 96/24 transfers to DVD-A is still breath taking. I can't wait to get the Dolby applied to it.

Before I start this, while I understand the technical concepts in some detail, I struggle with naming conventions. There are those of you out there who take great pleasure in pointing out silly things like "It's not a decoder its a ?????". I'm not a purist, I'm a hobbyist so please spare me the jargon correction. The entire system is a compander and it doesn't really matter what you call the pre- and post- processing. So I'll use the word "decode" for post-processing, even though it isn't strictly a decoder as it's all done in the analog domain and there really isn't and coding, per se. But I can't come up with a better word for it. Even if you have a better word, I'm not interested because I'm certainly not going to go back and fix all my posts once I get a better word and changing terminology midstream in a technical paper makes for unclear writing. But please, it's not that I don't want any input. I really really want it. Just don't be that guy in class who makes us all groan as you point out some useless bit of information that isn't on the test just for the joy of feeling smarter than everyone. If you really have input, go for it. But don't tell me its not a decoder, cuz I already know that.

Now, on with my post.

This website says it's an approximation, but even that would be good.

I've been thinking a lot about how I might do this digitally and it's one of the most difficult things to reproduce in the digital doman because the settings vary continuously, not in steps, in two different dimensions. Continuous is what makes analog so palpable and can only be approximated digitally. A proper digital Dolby filter/expander/decoder might take so much memory and processing that it becomes infeasible unless you make some compromises.

Reading everyone's posts, which are great, BTW, I'm not seeing anyone discussing the two dimensional nature of Dolby B. If you read the white paper on it, available at:


http://www.dolby.com/uploadedFiles/English...ion_Systems.pdf

You'll see that the equalization applied is varied by input frequency AND amplitude. The difficulty with simply applying a single equalization on decoding is that the curve changes in a continuous manner as amplitude increases and you'll either have to over or under equalize unless you address this. What that means sonically is very subtle, but it will either take the edge off or overemphasize the edge of some voices and other midrange content in the harmonics. It will also affect the relative volume of content in the equalized band as well, making a cymbal too loud or soft, maybe. These contribute greatly to the sense they are in the room, so figuring out how to do this would produce a non-trivial result.

So, a digital filter on foobar, winamp, steinberg or any other audio processing app would somehow have to evaluate amplitude and frequency and apply the filter. Not so easy. If you did it with a table, it would be an enormous table at 24 bits with total of 2 to 23rd curves for each amplitude. Furthermore at 96k the curves would each be 96k entries (I believe. this could be higher, but that's what the project is about)so you'd have s to 23 * 96,000 entries that would have to be in your plug in. Then it would be evaluated, looked up, then applied to the signal. ARGH! In analog it was easy, digitally, it's difficult.

I'm going to try this plug in and reprocess the content with this Dolby approximater and probably post another thread about it if anyone cares. I'm taking .AOB files, extracting .wavs, processing them through the decoder, then reencoding back to .AOB and burning the disc. It will be an adventure.

Thanks for all the posting about this topic

QUOTE (2Bdecided @ May 5 2009, 02:26) *
Tape Restore Live includes a Dolby B simulation.

It was programmed empirically, but it works well.

Cheers,
David.

Go to the top of the page
+Quote Post
Dynamic
post Jun 6 2013, 00:25
Post #8





Group: Members
Posts: 793
Joined: 17-September 06
Member No.: 35307



It's not just thin-ness and high end boost, but the decay tails of cymbals and similar high-end transients tends to last considerably longer thanks to the dynamic range compression applied to the high-end (which the expander part of the Dolby B decoder reverses). Applying Dolby B is likely to tighen up the decay in a lot of hi-hats, cymbals and similar sounds.

I guess that measurement of the hiss spectrum and level between tracks may also help to confirm whether Dolby B was turned on when digitized in comparison with known cases of Dolby B off and on (assuming the peak level of the loud audio is anywhere near full scale)

I've been using someone else's Fostex X-18 cassette recorder/player/mixer recently to digitize some old band recordings and backing tracks (it can use both stereo tracks on side 1 of the tape as well as both tracks on side 2 effectively playing backwards, to record 4 tracks of audio on one side only of a cassette over half the marked duration), and it has Dolby B which cannot be turned off, as far I'm aware. (By zeroing the faders on channels 3 and 4, it can also be used as a 2-sided tape deck).

I've previously used Tape Restore Live to reverse Dolby B encoding and adjust azimuth with good results, and used real Dolby B on my ancient Sony WM-36 Walkman (noted for decent reproduction when the EQ is set to zero, but lots of noise when non-zero)/ One of these days I must compare the WM-36 with Dolby B turned on to the same with Dolby OFF fed through Tape Restore Live.
Go to the top of the page
+Quote Post
Glenn Gundlach
post Jun 6 2013, 06:43
Post #9





Group: Members
Posts: 359
Joined: 19-April 08
From: LA
Member No.: 52914



I did a lot of tape machine alignment for use with Dolby B, both cassette decks of many manufacturers and open reel decks - mostly Sonys and Teacs used with Advent 100a or 101 Dolby processors. Dolby is quite good as long as the response of the deck is flat and the levels are matched. A slight hump or dip becomes a larger hump or dip - Dolby magnifies the errors. The only Dolby processors that were nasty were in early Teac cassette decks using early IC Dolby processors. Tone bursts ( which I used as 'markers' for measuring frequency response ) had a low frequency ripple - quite pronounced - that looked very much like an amplifier shifting its DC operating point from signal on / off. The discrete Dolby processors from Advent and Sony did not exhibit this fault. My Nakimichi 580M also does not have this anomaly but I don't remember if the Dolby is integrated or discrete. Sonically, it was a low frequency thump far below the turnover frequency when there was a mid / high frequency transient.

When I got to Hollywood I ran into Dolby A and then SR. Fortunately broadcast machines are better behaved than old consumer machines.

I agree the software decoder could be fun to try. Maybe when I'm off work for a bit with nothing to do.

G
Go to the top of the page
+Quote Post
2Bdecided
post Jun 6 2013, 10:50
Post #10


ReplayGain developer


Group: Developer
Posts: 4957
Joined: 5-November 01
From: Yorkshire, UK
Member No.: 409



@HelpfulDad - Tape Restore Live is certainly not a fixed EQ. Hope you find it useful. I think there are a couple of other threads discussing it on HA.

You seem to think that digital audio is much harder and less capable than it really is. Much harder problems have been solved, and using far less complexity. I agree that the design of Dolby NR processing is easy in the analogue world because that's how it was specified, but even if you did it digitally in the complex way you suggest, you could avoid such a large lookup table because errors of a fraction of a dB will be inaudible. Even 1 or 2dB wrong may be the least of your worries - calibrating the levels is critical, and the theoretical correct level doest not always give the best results, due to the signal coming from the tape not being identical to what was recorded.

btw, recording analogue tapes with absolutely no audible degradation at 16-bits 44.1kHz is also fairly easy.

Cheers,
David.
Go to the top of the page
+Quote Post
HelpfulDad
post Jun 7 2013, 14:19
Post #11





Group: Members
Posts: 3
Joined: 5-June 13
Member No.: 108504



Thanks David,

However, I beg to differ.

Bottom line about 44.1 is that it just isn't sampling often enough to capture multiple high frequency tones with phase offsets. It's a fallacy to think that just because you can produce a 20k tone reasonably well at 44.1k that the sound will be realistic in a real life setting.

If you had an audio wave form consisting of 5 15khz streams, offset in phase by 12 degrees, how could 44.1k possible extract all of the information from the wave form? It couldn't. There would be information missing between the samples. Hence the popularity of higher sample rate material. Similarly for 24 vs 16 bits. Even at 96k you'd be approximating the waveform more at 16 and 24. So, unless you have a library of single tone source material, properly executed faster sample rate and higher bit length is always going to more failthfully reproduce the origingal. Period. That doesn't mean it will sound better. Some stuff is terrible source material and mp3 makes it sound better, qualitatively to me. But, that doesn't change the limits at 44.1/16. Maybe you haven't heard good source material at 96k/24 bits. or 192k/24 It's stunning. Unbelieveable realistic. Even the dog sits up and listens. Again, all dependent on the source material and quality of the equipment used. jittery noisy stuff will make garbage.

And yes, you are right that I could approximate and hamfist it using big 1 or 2 db jumps to approximate, but if I'm going to do this, I'm going to try and be clever and it get so that it isn't an approximation like the winamp plug in apparently is. Like I said in previous post, its a hobby and challenge. cool.gif



QUOTE (2Bdecided @ Jun 6 2013, 01:50) *
@HelpfulDad - Tape Restore Live is certainly not a fixed EQ. Hope you find it useful. I think there are a couple of other threads discussing it on HA.

You seem to think that digital audio is much harder and less capable than it really is. Much harder problems have been solved, and using far less complexity. I agree that the design of Dolby NR processing is easy in the analogue world because that's how it was specified, but even if you did it digitally in the complex way you suggest, you could avoid such a large lookup table because errors of a fraction of a dB will be inaudible. Even 1 or 2dB wrong may be the least of your worries - calibrating the levels is critical, and the theoretical correct level doest not always give the best results, due to the signal coming from the tape not being identical to what was recorded.

btw, recording analogue tapes with absolutely no audible degradation at 16-bits 44.1kHz is also fairly easy.

Cheers,
David.

Go to the top of the page
+Quote Post
pdq
post Jun 7 2013, 14:39
Post #12





Group: Members
Posts: 3311
Joined: 1-September 05
From: SE Pennsylvania
Member No.: 24233



You really need to learn some of the theory behind this before you come out with a lot of absurd statements.
Go to the top of the page
+Quote Post
2Bdecided
post Jun 7 2013, 15:13
Post #13


ReplayGain developer


Group: Developer
Posts: 4957
Joined: 5-November 01
From: Yorkshire, UK
Member No.: 409



Oh HelpfulDad, you've really picked the wrong forum to post those kind of hand waving and incorrect attempts to justify high resolution audio. Your example is simply incorrect - if you actually did the experiment, rather than just talked about it and claimed "it can't work", you'd find that anything below 20kHz was preserved perfectly, irrespective of phase.

This is good...
http://xiph.org/video/vid2.shtml

You can avoid re-running an argument that was had several years ago by reading it here first...
http://www.hydrogenaudio.org/forums/index....7516#entry74075

Cheers,
David.
Go to the top of the page
+Quote Post
alanofoz
post Jun 8 2013, 00:20
Post #14





Group: Members
Posts: 46
Joined: 10-August 03
Member No.: 8294



QUOTE (HelpfulDad @ Jun 8 2013, 00:19) *
...There would be information missing between the samples...

That's been good for a laugh since the 1930s...


--------------------
Cheers,
Alan
Go to the top of the page
+Quote Post
mjb2006
post Jun 8 2013, 04:49
Post #15





Group: Members
Posts: 712
Joined: 12-May 06
From: Colorado, USA
Member No.: 30694



QUOTE (HelpfulDad @ Jun 7 2013, 07:19) *
Similarly for 24 vs 16 bits. Even at 96k you'd be approximating the waveform more at 16 and 24.

Another dubious claim, aside from those already called out above.

It's easy for us to say "go read up on this", but I find that a lot of the material to be found online either oversimplifies and ends up reinforcing misconceptions, or is technically accurate but assumes way too much of a math and engineering background. It's hard to find things that are in-between.

My suggestion is to go watch the two 30-minute videos at xiph.org, following along with the wiki ("Discuss and learn more" links). Maybe also review the discussion pages in that wiki, as there are some good questions and answers and corrections. If, after doing that, you remain unconvinced that higher sample rates are necessary for sub-Nyquist-frequency content, or higher bit depths are necessary for sub-6.02N+1.76 dB SNR (or whatever), either ask questions there or in a new thread here.

This post has been edited by mjb2006: Jun 8 2013, 05:09
Go to the top of the page
+Quote Post
saratoga
post Jun 8 2013, 06:20
Post #16





Group: Members
Posts: 4718
Joined: 2-September 02
Member No.: 3264



QUOTE (HelpfulDad @ Jun 7 2013, 09:19) *
If you had an audio wave form consisting of 5 15khz streams, offset in phase by 12 degrees, how could 44.1k possible extract all of the information from the wave form?


Theres exactly two orthogonal phases per cycle in your example, and Nyquist says >2 samples per cycle. So you have N unknowns and >N independent measurements. This is an overdetermined system. You can extract "all of the information" by just solving for the unknowns smile.gif

FWIW, wikipedia is actually pretty good in this instance:

http://en.wikipedia.org/wiki/Sound_quality

That then links to pretty much everything you could want. Worth taking an hour to read through.

Go to the top of the page
+Quote Post
Nessuno
post Jun 8 2013, 10:29
Post #17





Group: Members
Posts: 422
Joined: 16-December 10
From: Palermo
Member No.: 86562



QUOTE (mjb2006 @ Jun 8 2013, 05:49) *
It's easy for us to say "go read up on this", but I find that a lot of the material to be found online either oversimplifies and ends up reinforcing misconceptions, or is technically accurate but assumes way too much of a math and engineering background. It's hard to find things that are in-between.

I actually don't think that there could be an alternative way to the hardest one to completely understand sampling theory, which is undoubtably counterintuitive to someone who doesn't have enough background in math and information theory.
Teaching by simplifications and examples could be good, but there always be someone feeling smartass enough to come out with a "clever example none has ever considered before".
Behind a certain point, there's no real alternative to appeal to the "collective authority" of science and engineering. After all, how many people actually understand aircraft technology in full details? What's more counterintuitive? And still people fly! wink.gif


--------------------
... I live by long distance.
Go to the top of the page
+Quote Post
HelpfulDad
post Jun 8 2013, 14:26
Post #18





Group: Members
Posts: 3
Joined: 5-June 13
Member No.: 108504



I'm not trying to justify anything. Merely stating my own reasons for why I'm doing the work I'm going to do. The thread is about Dolby B simulation, so stick to topic.


This is good...
http://xiph.org/video/vid2.shtml

You can avoid re-running an argument that was had several years ago by reading it here first...
http://www.hydrogenaudio.org/forums/index....7516#entry74075

Cheers,
David.
[/quote]
Go to the top of the page
+Quote Post
[JAZ]
post Jun 8 2013, 15:03
Post #19





Group: Members
Posts: 1711
Joined: 24-June 02
From: Catalunya(Spain)
Member No.: 2383



@HelpfulDad: 2BDecided already gave you an adequate answer in his first reply, and showed you a very easy to understand video on why you have important misconceptions on what sampled signals are (which you've quoted in your last reply).

As such, I believe there's not much more to say about your interest in using this software simulation.

This post has been edited by [JAZ]: Jun 8 2013, 15:03
Go to the top of the page
+Quote Post
saratoga
post Jun 8 2013, 17:25
Post #20





Group: Members
Posts: 4718
Joined: 2-September 02
Member No.: 3264



QUOTE (HelpfulDad @ Jun 8 2013, 09:26) *
Merely stating my own reasons for why I'm doing the work I'm going to do. The thread is about Dolby B simulation, so stick to topic.


The problem is that if you do not understand what is being discussed, you will not be able to participate in this thread. I suggest taking a look at the resources provided. Its important that you understand the basics before trying to understand more complex topics.
Go to the top of the page
+Quote Post
2Bdecided
post Jun 13 2013, 14:41
Post #21


ReplayGain developer


Group: Developer
Posts: 4957
Joined: 5-November 01
From: Yorkshire, UK
Member No.: 409



Highly relevant to this thread is the fact that 16/44.1kHz is more than enough. There is no conceivable argument on earth for using more when capturing a cassette*. I don't even need to bring audibility into the discussion. If Tape Restore Live doesn't do the job well enough, it's not because the source material is sampled at 44.1kHz.

* - well, apart from the Plangent Process - but that still adheres to Nyquist!

Cheers,
David.

EDIT: Since HelpfulDad PM'd a multi-paged essay to me explaining how wrong and insulting I am, I thought I'd clarify in-thread...

QUOTE (HelpfulDad @ Jun 7 2013, 14:19) *
If you had an audio wave form consisting of 5 15khz streams, offset in phase by 12 degrees, how could 44.1k possible extract all of the information from the wave form? It couldn't. There would be information missing between the samples.
If you add 5 15kHz waveforms together with 12 degree phase offsets between them, you get one 15kHz waveform, with a 13.58dB higher amplitude than the originals, offset 24 degrees from the first and last of the originals (i.e. equal to the third/middle of the original waveforms). The component waveforms, their addition, and the final waveforms, can all be perfectly represented at 44.1kHz. You can try the experiment in most audio editors, but I'll save you the trouble: it works equally well whether you set the sampling rate to 44.1kHz or any other higher value.

I don't have a particular love for any sample rate, but attributing certain problems to the choice of sample rate, and then attempting to solve them by increasing the sample rate, will not solve any problems which are not caused by the choice of sample rate.

This is such an obvious logical point that it's hard for me to see how someone can be offended by the attempt to determine whether a certain problem (a) exists, and (b) is due to sample rate.

Cheers,
David.


This post has been edited by 2Bdecided: Jun 13 2013, 16:42
Go to the top of the page
+Quote Post

Reply to this topicStart new topic
1 User(s) are reading this topic (1 Guests and 0 Anonymous Users)
0 Members:

 



RSS Lo-Fi Version Time is now: 23rd April 2014 - 18:58