How to clean noisy sound from VHS capture?

[Darkwhisperer]

I've captured some old VHS tapes onto my PC, & am going to convert them to DVD. However, the sound aint too great, its pretty noisy, & theres a heavy hum in the background.:(

The little software that I've found which claims to clean audio is pretty expensive(especially in South African Rands).

Does anyone know of any free software, or perhaps plug-ins, that can help me? Or any other way that can help me?

Thanks! :)

[AVIL]

Hi,

You can demux the sound with virtualdub/virtualdubmod and denoise it with audacity (free software from sourceforge). Trick is select a short period with only noise (the start of the tape, a silence, ...) and tell audacity that erase this noise. Default values usually perform well.

There are other programs (e.g. goldwave) more easy to learn, but is shareware. The method is the same or using some presets (hiss reduction). The trial mode (free) was almost fully featured.

After, I advice you to normalize the sound and, of course, remux (virtualdubmod is very suitable)


Good luck.

[Darkwhisperer]

Cool! I'll give it a shot & let your'll know!

[Mug Funky]

i use audition (formerly cool edit pro 2, and not really improved at all thankyouverymuchadobe) for that stuff, but i imagine there's better stuff out there.

here's a hint: lowpassing is a very good way to reduce noise without hurting the sound - optical film tracks rarely have anything over 8000 hz, so you can often just lowpass to that and 2/3 of the noise is gone. the rest can often be handled with multi-band dynamic processing (like a noise gate, but working on several bands).

FFT based NR methods are good, but leave far too many artefacts which are almost always more objectionable than the noise you were reducing :). the trick is remove only as much as you need to, and leave the rest in to mask the artefacts introduced.

[mg262]

Darkwhisperer,

I've used Audition too, and I found it startlingly effective on tape hiss from my old VHS tapes. When I started I didn't really notice any artefacts, but now I sometimes notice a slight metallicness/tinniness. It should be possible to remove the hum almost completely if it's fixed on a particular frequency... though I don't know much about this stuff myself. I would also recommend trying the click/pop eliminator if you are working with material this bad; I've found it very effective on high generation material, and I didn't notice any resultant artefacts.

Mug Funky's suggestions are excellent and I'm going to try them myself!

[Mug Funky]

about click/pop and NR:

do the click/pop first, as the noise reduction will skip over transients and leave a little echo before and after them as well. this can be quite marked on big clicks.

also do NR on 32 bit float only - it will demolish any 16 bit sources. and only run NR once for the same reason.

one thing that's very useful is the little frequency vs NR amount graph. it's buggy, but very useful (it will artefact strangely when you set NR amount for 1 frequency less than 50%, but above 50% it's fine). use it at 100% above 8000 hz or so, and below about 70 Hz - this will kill rumbles and pilot tones very nicely, and if there's any actual signal around it'll pass it through reasonably well. as for the above 8000 range, be careful. it can also help to have the least amount of reduction in the 1000-3000hz range, as this is where voice likes to hang out. previewing with "keep only noise" on helps you hear if any voice is being wrongly removed (you'll be able to hear raspy speech in the noise. a little bit is inevitable, but keep it to a minimum).

damn... i've spent way too much time doing this stuff. and in the end, i wish there were better tools out there. i can think of several ways to improve it, but haven't the maths or inclination (or faith enough in these ideas) to implement it :)

@ clouded:

if you are interested, i can send you a .ses file that does the multiband compander thing already - just drop in the lightly FFT denoised clip and you'll get a rather transparent extra noise reduction at the cost of losing a bit more signal. EQ'ing is also pretty necessary doing it this way, but you'll want to do that anyway to keep it all sounding consistent. multitrack mode is very useful for stacking effects, especially several dynamic effects in a row.

[Boulder]

Mug: would your methods be effective for all kinds of old sources such as C-cassettes? I've been thinking about finally restoring some old tapes from the early 80s before they break down as soon as I have a couple of days of spare time.. I've also been wondering what sample rate I should use for digitizing them - I'd appreciate it if you could give a hint or two. Thanks.

[mg262]

if you are interested, i can send you a .ses file that does the multiband compander thing already - just drop in the lightly FFT denoised clip and you'll get a rather transparent extra noise reduction at the cost of losing a bit more signal.Absolutely -- thank you very much!

[AVIL]

@boulder,

I digitize at 44100 Hz. My card admits 48000 Hz but some software (e.g. CD burn) refuses that rate. As my tape deck barely gives up to 15000 Hz this sample rate is enough. I think 8000 Hz cut is very radical for casette music. I use FFT techniques to denoise but sometimes (when too frequencies are eliminated) another interference noise (even worse) arises.

Another tech is used in d-noise software. It uses a sound coherence approach. And enhance results with iterative procedures. You caun found it at:

http://www.fmah.com/AUDIO/AUDIODEN.HTM

This last version is rather old (for Windows 95) but runs in windows XP (the wave display not).

The iterative approach consists in re-filter the sound discarded to obtain clean sound that is added at the previous obtainded clean sound). I never obtain clean sound in the fourth iteration. And the sound obtained in third iteration is almost negligeable.

[zilog jones]

Mug: would your methods be effective for all kinds of old sources such as C-cassettes? I've been thinking about finally restoring some old tapes from the early 80s before they break down as soon as I have a couple of days of spare time.. I've also been wondering what sample rate I should use for digitizing them - I'd appreciate it if you could give a hint or two. Thanks.
I've tried the NR stuff on Audition with old VHS audio and it helped greatly too. I'd recommend using the highest sampling rate you can when recording them on your PC. IIRC, with professional studio recordings they use 48kHz for the masters then downconvert them to 44.1kHz for CD production.

[guada 2]

Hello :) ,

Adobe Audition is very good with Bias Soundsoap Pro plug in.
But I prefer "MAGIC SAMPLITUDE 7" for the optimal settings with Waves plug in.

[Mug Funky]

for sample rate, recording wise (so i've heard) precision v samplerate tradeoff is best at 24/48. you wont get anything useful above that, especially on old and degraded material.

for processing, the story is different - there's no physical limitations on a computer, so you can work in 96k (but there's not much point on such material either. it'll just slow down and make it harder to work with). i just did all my stuff at 48k because that's what it was recorded in and that's what it would be presented in (DVD).

@ AVIL: yeah, i wouldn't recommend 8k cutoff for cassette stuff, as it's got a way higher bandwidth. remember i was working with optical film tracks that had been recorded on VHS. anything that was came from higher quality masters will need more careful treatment.

@ all: ALWAYS keep your original recordings. the NR technology is changing all the time, and i might be talking crap, so a copy of your original recording can really help. DVDs are cheap, CDs too, so there's no reason not to back it all up in case you come up with something better. remember that all NR removes signal as well.

[Boulder]

I'm planning on making regular audio CDs out of the tapes so I guess I'll sample at 44.1kHz. I was just wondering whether I should sample at 22kHz or 32kHz, do NR and then use SSRC to resample to the final samplerate.

[Mug Funky]

i'd still probably say 48k, but it's a tough call to make. depends how good the A/D conversion is (44.1k requires a steeper brickwall filter which means good 44.1k A/D converters are more expensive to make than good 48k ones. a good software resampler may or may not be a better option... but the audible difference would be bugger-all).

if you want to get through this in less time, then 44.1 is the way to go :)

[edit]

@ clouded:

i'll have to make a zipped (rar'd, 7z'd, whatever) package for you with some docs and a couple of impulse files that can be handy... hope you don't mind a bit of a wait on that .ses file... tweaking is extremely necessary, and though it's quite simple, it helps to know exactly what needs to be tweaked :) right now it's set for a VHS sourced optical 16mm film track, and even then massive tweaking is usually needed per-episode.

[mg262]

Honestly, there is no hurry for it -- audio noise reduction is something I'll use for final version DVDs, and I won't make those until I've finished Cel! So you can see that it is not a pressing need...

[BruceL]

I have used various versions of "Audio Cleaning Lab" over the years for processing Broadcast capture, VHS capture, audio cassette transfers and Album transfers with good results. Includes many tools as shown here: http://site.magix.net/english-us/home/music/audio-cleaning-lab-10/acl-10/new-functions/?no_cache=1&version=standard
I have purchased versions over the years at CompUSA or Amazon and either got for free or $9.95 max after rebate. Hope someone finds this useful.

[Boulder]

i'd still probably say 48k, but it's a tough call to make. depends how good the A/D conversion is (44.1k requires a steeper brickwall filter which means good 44.1k A/D converters are more expensive to make than good 48k ones. a good software resampler may or may not be a better option... but the audible difference would be bugger-all).

if you want to get through this in less time, then 44.1 is the way to go :)

OK then, one more question (which nicely suits the topic as well ;)) : when you do a 16bit->32bit->NR->16bit chain, do you use dithering when converting from 32bit to 16? If so, what kind of settings are recommended? I have to ask this as I've never done it, I've always processed the audio track in 16bits.

[codeguru]

Hmmm, I think it doesn't really make any difference (samplerate conversion, 16 vs 32 bit) if you have worse tape sound (noisy, weak, freq range 100hz to 8 khz).

It is true that 32 bit is more precise but remember, a 80'ies VHS tape has barely more than 45 db Noise/signal ratio, and 48 db is equal to the dynamic range of 8 Bits. That means there are 8, maybe 10 bit of information in the audio signal, but not more. If you loose 2 bits due to integer processing you won't really loose anything because you have your 10 bit of information in the remaining 14 bits.

Resampling already occurs when you have tapes with bad timings (Virtual VCR)..... you won't notice any difference, maybe you will hear it if you sample HQ material, EG. laser disc or movies with Hifi sound (usually has 90 db or better)

If you get echo-like artifacts with Audition or Cooledit with the denoising the S/N ratio is too worse. I tried this with a movie that had only 30 db :-( of audio and got strange audio artifacts. But this tape was limited to frequencies up to 5 khz so I did a 5 khz lowpass, 100 hz highpass filter and made the denoising with 50% filter strength.

[mod]

If you want buy FLStudio.. it's not made for that, but I use it also for that and the results are great..