So I’ve been getting tunes ready for RMAF. I’ll be bringing both burned compilation CDs with RB and a USB stick with RB, High Res, and DSF files. What I discovered by some surprise is that my high res tunes from Linn sound absolutely odd when converted to RB for CD burning using dBpoweramp. There’s a weird hashy noise floor that you can pick out quite easily on every Linn tune I converted. Neither the Blue Coast HR tunes nor others produce this same sort of noise. I’ve converted many a HR tune to standard RB for CD burning. Linn tunes seem to be the only ones that have this oddity.
Particularly odd as Linn’s downloads do not include DRM.
Have you tried anything other dBpoweramp to down sample?
Do the files on Linn’s test downloads page exhibit this? If so, which ones?
I just downloaded and converted Linn’s 24bit 192kHz Free Lossless Audio Codec test file using a couple of different software packages, including dBpoewramp. No issues
It is as revealing a source as one can get, solo clarinet and piano.
It also happens to be a great example of why small scale dynamic retrieval is critical. A classical clarinet is played without vibrato. Thus, their primary means of expression other than rubato is the use of dynamic shadings.
Doesn’t happen with the test file. The Linn HR files I purchased sound great when played back as HR files, but many (it turns out, not all) sound weird when Redbooked. Edgar Knecht’s “Fruhling” is a good example. So is (ironically) Katherine Bryan playing Flute Concerto 1993 from the Rouse & Ibert download.
Have you tried anything other dBpoweramp to downsample? I am more suspicious of dBpoweramp than of the files themselves. My guess it is an aliasing error.
Do you have the ability to create a brief cut of an original file which does this, and of the resulting noisy file, zip them and attach to a post? We only need ten seconds or so.
I hope this works. I’ve used other versions of dBpoweramp over time, but I’ve done other HR to RB conversions with this version (many at the same HR rates - 192/24 or 96/24) with no issues.
Hmm, won’t let me add WAV files. I’ll email them to you.
If you zip them you should be able to attach it to the post.
Here are the files he sent, plus a version I converted (44-16) with iZotope Advanced.
Yipes Fruhling_copy is just “noised” (Green)
Fruhling 44-16 (iZotope Advanced) is still a little “noised” (Blue)
I built Fruhling_Audition with Adobe Audition 3.0 and it is correct (Red)
The original sample is interesting in that it has some significant peaks above 22k (bottom picture)
Original:
Thank you, Ted! Very, very cool!
If you would be so kind, please create a spectrum of these two files. The first is 44-24; the second 44-16 with no dither (if I set things correctly).
I am curious as to whether the noise comes from the dither I applied when truncating to 16-bit.
Also, what are your spectrogram settings? I look primarily for noise peaks and the like when cleaning up recordings, but have not found settings to be as useful as yours for looking at this sort of thing.
Your two files look just alike and both look like my Audition file (well, with the exception that I used the prefilter in Audition which does the steep rolloff just before the Nyquist frequency.)
For more detail I always set the FFT size as large as I can get it (in this case 65536.) To make sure that there are 65535 samples before the first sample I select from approximate two seconds up to 2 seconds before the end. (In this case I selected from 2 seconds to 8 seconds.
For the sharpest envelopes around each frequency I window with the steepest window available (in this case Kaiser 180)
Then I do a scan over the selection to average out the spectrum.
I did some more playing with my software and came up with spectrum analyzer settings which answer my questions.
The increase in noise above 16kHz in Fruhling 44-16 (iZotope Advanced) is indeed the dither I applied when truncating from 24-bit to 16-bit (MBIT+ which is one of the dithers recording engineers go nutty over).
My latest versions (44-24, and 44-16 no dither) look like Ted’s red spectrum, complete with the odd little peak at 19.3kHz or so.
The good news is dBpoweramp is indeed the culprit as I suspected. The bad news is dBpoweramp is doing a horrible job of downsampling.
If you use TPDF (triangular) dither you get a spectrum that’s indistinguishable (at the scale above) for 16/44.1 and 24/44.1 That’s why I didn’t post it the first time.
Thanks, Ted!
Great stuff
Actually it isn’t dBpoweramp that’s at fault.
I used dBpoweramp and got a spectrum that’s identical to the Adobe Audition 3.0.
You need to use the DSP effects to put in a resample to 44.1k and then a DSP effect to do triangle dither to 16 bits.
[Edit:]
Duh, you don’t need the DSPs. Just doing the wav convert with the output at 44.1k and the bits at 16 works well. (I’m using release 15.1)
I’ve got no idea what went wrong in the first place - perhaps explicitly using the DSP first sets the default dither to triangular?
Ah ha! I am NOT using 15.1. I just downloaded that last night, suspecting it might be due to an earlier version bug. Yes, I normally just select the WAV convert option at 16/44.1, nothing special. Are you saying that comes out the same as DSP with triangle dither?
Gah! This means I’ll have to reburn my CDs for RMAF. I have only two TDK T-Y 74 minute CD-Rs left. After that they’re gone! I’ve discovered the 80 minute T-Y CD-Rs do not burn with the same low order of C1 errors and Jitter, and they don’t sound as good as the same tracks made on the older 74 minute jobs. I’ve tried every recording strategy I have available from my Plextor Professional XL burn software and they just do not come out as well as the old discs. (Yes, I know C1 errors in theory should not matter, but I’ve observed that all things being equal the CDs that to me sound the best also have the lowest C1 and Jitter numbers - and yes I discovered this after I learned these things could be measured. My ears told me before I measured.)
ARGH!
We feel your pain.
But again, great ears!
tony22 said Ah ha! I am NOT using 15.1. I just downloaded that last night, suspecting it might be due to an earlier version bug. Yes, I normally just select the WAV convert option at 16/44.1, nothing special. Are you saying that comes out the same as DSP with triangle dither?I knew that I'd have the most control when using explicit DSPs so that's what I did first and got good results. Later I thought that I'd better not condemn the most obvious use of dBpoweramp without explicitly checking it. With the rest of this thread I was mildly surprised when it got the exact same result. I don't know if anything affects the type of dither using the simple 16/44.1 conversion without DSPs, just that I apparently got TPDF when I did it :) Spoon is good about answering questions at the dBpoweramp web site, but I didn't feel like plowing thru the posts for a precise answer.
I’ll be happy to check a few conversions for you if you want to experiment, tho your ears seem to be working fine
Ted Smith said Actually it isn't dBpoweramp that's at fault.I used dBpoweramp and got a spectrum that’s identical to the Adobe Audition 3.0.
You need to use the DSP effects to put in a resample to 44.1k and then a DSP effect to do triangle dither to 16 bits.
[Edit:]
Duh, you don’t need the DSPs. Just doing the wav convert with the output at 44.1k and the bits at 16 works well. (I’m using release 15.1)
I’ve got no idea what went wrong in the first place - perhaps explicitly using the DSP first sets the default dither to triangular?
Turns out you do need the DSPs. I tried it with 15.1 with just a straight 16/44.1 WAV conversion and it did the same thing. When I did as you suggested it converted fine.
Do you think it’s worth leaving even for 16/44.1 standard FLAC conversions? Will it hurt those?
When you write “standard FLAC conversions,” do you mean converting a 44/16 WAV to a 44/16 FLAC?
If so, this is unlikely to exhibit a similar problem as there is no reason to add dither.
If you would like, post a snippet of an original file and the snippet after converted. We can then take a look to see if anything untoward occurred.
The other way 'round, Elk. FLAC to WAV so I can make a CD-R. But after I posted I figured the same thing. Why do unnecessary resampling or dithering?