How to check recording quality

[Mike in MD]

Recently, wgscott showed me how to generate an FFT spectrum for the hiRez files that we listen to. I had spent a few months searching for a program that could do this for under $100. It turns out that there is a free program that does these spectra: Audacity.

 

I am not aware of a thread on CA that explains how to do these FFT spectra, so I am adding this tutorial. But first I would like to show you some interesting results.

 

This first spectrum is:

Linn, Dunedin Consort, JS Bach, B-minor Mass

 

One. (I don't know how to embed the screen shots into the body of the text, the way wgscott did it. The files are listed below as attachments.) (But the initial group of attachments were dropped, so please look at the second comment which now has the attachments.) This screenshot is "Linn Bach."

 

Look at the clean highs stretching up to 40k Hz. Great recording production. And great music. I was just listening to it.

 

Congratulations Linn !!

 

Then there are some that didn't match this high standard.

 

I know that Blue Coast Records has a goal to put out great recordings, but look at this.

Blue Coast, Tidbits, Art and Bruce, Impressions of Duke, Satin Doll

 

Two. See "Blue Coast" in 2nd comment.

 

The high frequencies are just filled with noise. Something is wrong with the recording equipment.

 

Next is (unfortunately) one of my favorite tracks. Most Chesky hiRez is good, but.... (and I know their download manager and customer service are terrible. I've had problems, too.)

Chesky, Ron Carter and Rosa Passos, Entre Amigos, Bahia

 

Three. See "Chesky Bahia"

 

There are two noise bands at 16k and 32k Hz running right across the track. How does that help the music???

 

And I thought that, with 24 bits for amplitude data, clipping would never happen. But look at

Chesky, Jimmy Cobb, Cobb's Corner, Book's Bossa

 

Four. See "Chesky Jimmy Cobb"

 

The red vertical lines are clipping, according to Audacity. With 24 bits to work with, this should not happen. I know it is only three times in the track, but that is three too many.

 

I think that, if we check the recordings that we buy, then we can get the music producers to improve their albums.

To break this up a little, I will put the instructions for Audacity in the next "comment."

 

EDIT: the attachments disappeared from this post. See the reply, two down, for all the attachments (screenshots).

 

[Mike in MD]

First go to

http://audacity.sourceforge.net/

 

Download version 1.3.12 Beta. I tried 1.2.6, but it was not reliable for FLAC files. Audacity is free, but you can donate at the Get Involved tab, then choose Donate. I did. This is a good program and will help us improve the hiRez music that we listen to but letting the producers know that we are checking their work. (repeating myself.)

 

I run it on XP. They say it will work with Windows 7, XP and Vista and Mac, including OS X 10.6

 

When you have it installed, open Audacity. Go to: View -> Show Clipping. Click on Show Clipping. It will bring up a check mark. Now when clipping is detected, red lines will be inserted in the waveform display.

 

Go to: Edit -> Preferences (at the bottom of the menu). Select "Spectrograms." I have been using min freq = 1000, max freq = 40000, window size = 2048.

The window size is associated with the FFT program. A larger window is slower. These choices seem okay.

 

Now go to: File -> Open. Navigate to a WAV, AIFF or FLAC file. They support other formats, too. Select and click Open.

 

The first display is the Waveform, which shows sound amplitude. It also shows the red clipping mark, if clipping is detected.

 

To see the frequency spectrum, click on the inverted triangle on the left side of the spectrum, which I tried to circle on the screenshot below.

 

When you click on the triangle, a menu drops down. Click on Spectrum. Then, count slowly to fifteen. Audacity is slow here. And there is no indication that it is working on anything. Just wait.

 

You are now looking at the frequency distribution for the track. You can see if it really is a hiRez recording. And also can see any problems that were created in the production process.

 

If you want to save a PNG screenshot, like above, go to: Help -> Screenshot Tools. In the dialog box, first, choose the folder in which to save the images. Then I have clicked on "Capture Window Only." This gives you the Audacity whole screen. In the next section of choices, you can choose "Track Panel" and it will save only the spectrum portion of the window. Audacity does not tell you that the operation is done. You have to look in your folder.

 

When you Close the file or Exit from Audacity, the program will ask if you want to save your changes. Definitely choose NO. Audacity is an editing program. Unless you are deliberately editing the hiRez file, you don't want to save anything.

 

I think this is everything. It can only help if we can check on the quality of the recording process.

 

This description is based on an entire week of experience with Audacity (tongue firmly in cheek). But I am very impressed with its ease of use. Please add descriptions of the interesting features that I have missed. I would still like to know how to tell if the recording uses over 16 bits of amplitude data. This is different from the clipping issue.

 

 

[Mike in MD]

the first set of attachments seem to have disappeared.

I will try again.

 

 

[GeorgeW1238]

Mike,

Thank you for this very informative set of inputs !

I am just getting into digital audio at home and the availability of material that I want to listen to and the quality of it has been a concern to me.

I have Audacity, Media Monky, Foobar 2000, Sound Forge 10, and Media Center 15 installed so that I can evaluate their capabilities before loading up TB's of media.

I tried the Audacity spectrum analysis tool across MPEG3 files, WAV 44.1/16 files, WAV 48/16 files recorded from Direct to disc records,WAV 96/24 files recorded from records and 3 different HD Tracks at 96/24.

In my analysis I have observed the following :

1) MPEG3 - None of the clipping, noise,and limited to 22KHZ

2) WAV 48/16 from records - same as above

3) WAV 96/24 from records - same as above

4) HD Tracks FLAC #1 - NO clipping, NO noise bands, 22Khz limited

5) HD Tracks FLAC#2 - No clipping , band of energy (noise?) at 28KHZ and output to at least 40KHZ

6) HD Tracks FLAC #3 - No clipping, noise bands, and output to at least 40KHZ ( aka - looks like HIGH REZ )

I then listened to the files with the following observations :

1) Sounded lifeless - would never listen to this !

2) Sounded clean maybe a little lacking in dynamic range and " sparkle "

3) Same as above - wasted storage space !!

4) I think this was an upsampled CD - didn't sound as clean as my 44.1 & 48/16 files.

5) Sounded very natural, I could listen to this. Is the " noise band " being filtered out or enough above my hearing ?

6) Also Sounded natural - about the same as SACD material

 

My conclusions :

Spectrum analysis is an interesting tool but it doesn't tell the complete story at the level displayed with Audacity.

The high rez material available is of ? quality.

High Quality - High Rez files can sound excellent.

 

Thoughts :

Given the Spotted history of DVD-A & SACD - High REZ audio servers are a natural evolution for technical savvy users who enjoy high quality music.

 

BTW - Foobar 2000 also as a spectrum analysis function but it is not quite as usable as Audacity !

 

 

 

 

[wgscott]

This is John Coltrane, Lush Life, from HD Tracks, purchased a couple of weeks before they yanked it without explanation. It is supposedly 96kHz sampled, which means it should go to 48 kHz. You can see it abruptly truncates at about 22 kHz.

 

 

[cookiemarenco]

Hey Mike, I just saw your post. I must give you credit for searching through the Blue Coast online 'catalogue' and finding the Art Lande and Bruce Williamson recordings. I'm curious why you chose those recordings to be your first to analyze.

 

I do have a comment. Since I'm not sure what the analysis program is reading and what you're hearing, I'd like to point out that all the performances of Art and Bruce were played incredibly soft. What you might be hearing as 'noise' is actually the sound of air blowing through Bruce's saxophone, along with some 'spit and grit' through his reed. I'm not sure the analysis can detect the difference of air and extraneous noise. It happens I like guitar noise, air in a horn player's notes and pedaling sounds that come from a piano. If it's real, I leave it in. EQ'ing out is a mistake, in my book.

 

I think it's unfair to compare recordings without consideration to context. It happened I was recording Bruce's full length album that week when I asked Bruce and Art to perform a couple of pieces for Blue Coast at the last minute. Purely for my pleasure. I didn't have much time before they had to leave for the airport, and truthfully, I erred on the side of getting them playing than trying to optimize the sound. As a producer, I will ALWAYS choose a great performance over perfecting the sound. We offer full length previews for anyone to listen to if someone doesn't want to spend money on it.

 

I find that pictures of waves are far less accurate than human ears. Waves can't distinguish spit from a digital glitch. Human ears go beyond the scope of pictures of sound. I'm not a fan of looking at music and prefer to record on analog tape. That's just me.

 

At RMAF I put on a few listening seminars doing blindfold tests that compared the same SACD mastering with a CD gold disc mastering. The source was reliable because I was the mastering engineer. This is a very difficult test, but everyone could hear a very slight difference, the SACD winning out everytime.

 

We tested an anti-static 'pocket' for CDs that made an incredible difference in the high end playback and harmonic distortion. I would be curious if you could do a wav analysis on those kinds of differences and what your results show. Or if an analysis would show the difference in a digital copy or a downsampled 96 to 48 piece of music. That could be of some value.

 

I do agree that raising the level of recording is a good idea.... raising the level of musicianship is a better idea. Having a budget to record at the highest level possible with great musicians ALL the time is the BEST idea... I'm starting my non profit next week and donations are welcome!

 

By the way, I record to analog tape as my multitrack source, but, if great music was being played and I grabbed it on my cell phone, I'd release it... no one has to buy it, I just have to love it.

 

thanks!

Cookie

 

 

 

 

[cookiemarenco]

Hey all, I didn't want to give the impression I'm totally against wave viewing tests, eventhough I don't believe they are as accurate as hearing tests.

 

Here is a case where we used a wave form comparison to help us with our distributor. We recently released the Blue Coast Collection as a CD Gold disc, at the request of our distributor.

http://bluecoastrecords.com/collectorsedition

and the wav forms

http://bluecoastrecords.blogspot.com/2010/05/blue-coast-collection-collectors-gold.html

 

We chose to remove the compression used in the original mastering session of the CD layer on the hybrid disc. The Gold Disc now is the same mastering session as the SACD 2 track layer. Our distributor thought something was wrong with the disc because it was 6-7 db softer than the original CD with compression. He was going to reject the shipment until I sent in the comparison of the wave forms and showed him the 'flat topping' that occurs with the compressed CD disc vs the 3 peaks that happened in the entire Gold disc.

 

I asked him to use his volume knob and TURN IT UP.... and he did... but, what really sold him was the wave form comparison. At first, I thought it was silly to prove the point in a picture, but, it's ended up in countless reviews to our favor.

 

Many of us hear the difference in a FLAC and delivering full uncompressed Wav. This is such a subtle difference, I don't think Audacity will show it, but, I'll bet I can teach folks to hear it. I spent last week comparing DFF files played back on Foobar, vs Audiogate and DSD files burned to DVD played back on the Sony playstation. If I trusted an analysis program to do what my ears can do, I would GLADLY welcome them.

 

I don't think the analysis is good enough to tell the difference between a Steinway, Bosendorfer, or Yamaha piano...... but human ears can. Taking music out of context without knowing the source recording format, conditions, etc might suggest Miles Davis should never have released "Bitches Brew" because there's distortion on keyboards. Does anyone care about the low frequency pops from Nat King Cole's singing on his great recordings? Capturing a great performance does not always go hand in hand with a great recording.

 

Sorry to ramble about this, but I think we need to be fair to all those projects being 'looked' at vs 'listened' to.

 

thanks for understanding my point

Cookie

 

 

[Mike in MD]

Hi Cookie,

 

You are right... The hi freq component could be spit and reed noise. That possibility did not occur to me.

 

I looked at the spectrum for the Blue Coast tracks because I had recently downloaded the files and they were still on my computer. I have two computers: one for downloading, then a second for a music server. When I run out of space on the download machine I erase files. The Blue Coast tracks were still there.

 

I put the spectrum here because I was surprised to see the hi freq haze. The samples from the "raw" CAS music that you recorded are really clean. I was listening to those Blue Coast tracks yesterday. Good music. And free. I recommend that CA members listen to them. (Am I forgiven?)

 

If you follow the steps for running Audacity, you can easily see the track spectra yourself. This is a new tool that I am trying out. And I still don't understand the noise at 16k and 32k in Chesky Bahia.

 

Mike

 

 

[cookiemarenco]

Hey Mike, no worries. I understand you're interested to know more and I'm here to help when I can. Thanks for the thumbs up on the CAS tracks. The live recording chain we used gets no cleaner than that. I don't consider it my best work, because there is no second stage mix involved.

 

I was under headphones about 8 feet from the musicians and it was tough to mix where I liked it. Also, the room was very dry and I prefer a little more 'reflection' to the sound. Remixing will require the use of effects which bring up additional room tone or electrical sound... thought quite minimal. The music is great, though, so we made them available for free until I mix... which may be never at the pace I'm going.

 

Of note... those CAS recordings are very clean, however, I wonder if you listen carefully, you can hear the air conditioner in the background. We couldn't turn it off on ... just another peril of recording in a hotel room....

 

I'd like to hear Bahia and see if I can figure out what the noise is that is showing up. Similar to an airconditioner that you have no control over, there are a couple of things that come to mind.

*A TV monitor in the 'on' position will give off a 16-18Khz sound.

*A video camera could give off a buzz.

*A power drop or surge through a system can deliver a soft buzz/sound for hours during a recording.

*Fluorescent lights or inexpensive dimmers can create a sound in that range.

*Computers plugged into the same wall sockets can sometimes create that problem for remote or studio recordings.

*Air filters in the off position but plugged in can create a slight buzz.

*Sometimes electrical events can create sounds outside of the hearing range that will show up on either end of the spectrum.

 

Recording engineers deal with these unexpected issues all the time and learn to to work around them. Sometimes, you can't cancel a session. If you rent Grace Cathedral and it rains, you get rain on your recordings... no refunds for rain. And the cable car can't be taken out of the best performances of a live session at the Cathedral.

 

oh well....

 

we keep trying...

 

cookie

 

 

 

[bottlerocket]

This is 24/96 Susan Tedeschi from HDTracks. I see the flat top at 22 khz but also see very limted extensions beyond it. To me this looks completely compressed.

 

 

 

[Mike in MD]

That spectrum looks like up-sampled 16/44. Good 24/96 does not have that shape. I just had a bad download from Linn that I posted under "Look at what Linn sold as 24/96." I am going to check all my so-called HiRez files.

 

Mike

 

 

[Miska]

Two. See "Blue Coast" in 2nd comment.

The high frequencies are just filled with noise. Something is wrong with the recording equipment.

 

I don't think it has anything specifically wrong. My vague guess; looks like a DSD(64) recording either made with old'ish DSD gear and/or converted to PCM with less than perfect conversion. No strong ultrasonics coming form the mics.

 

With modern DSD64 gear and nice PCM conversion the noise floor should be flat (under -120 dBFS) until 20 kHz and there onwards rising smoothly. For DSD128 the noise floor should start rising from 40 kHz. Typical ultrasonics should smoothly disappear in the noise by 50 kHz in DSD64 and by 100 kHz in DSD128, crash cymbals and such could go higher but are generally harder to distinguish graphically from noise.

 

In other words, looks like a recording best listened in original DSD form...

 

I've attached a spectrum of one BlueCoast sample I converted from DSD to PCM myself. Note that the strongest spectrum peaks go to over 50 kHz.

 

[bottlerocket]

The S. Tedeschi example I posted is no longer on HDtracks. That tells me a lot. I guess they owe me $6.

 

 

 

[bottlerocket]

Oh no, Donald Fagen Nightfly DVDA looks upsampled too. I'm crushed. I never did hear much new on this versus the CD. Same with the Santana Supernatural DVDA. It sounds compressed and it is.

 

 

 

 

 

 

 

 

[Synfreak]

Hi!

 

Please note that material which was recorded (multitrack) or mastered in 44,1khz/16bit at the first time, as for example the Donald Fagen Nightfly album, you never ever will get sonics behind the 20-22khz mark.

 

But it may be beneficial to (re-)master these files at higher samplingrates.

 

The same applies to a lot of recordings made from the very late seventies up to the mid (or even end) of the nineties (and sometimes even later).

 

But actual recordings which are sold as "Hi-Resolution" of course should be so (but most of the time they are not...).

 

Cheers

Harald

 

[RayW]

I took a look at the spectra of 6 of my purchases from HD Tracks, specifically to determine what exists above 22 kHz. (I used Adobe Audition to do this.) I'm disappointed to report that only 1 of the 6 seems to have musical information there: See attached Phil_Orch_Strauss_Hindemith.JPG.

 

Two recordings in my sample look like CHSA_5048.JPG. Here there is rising quantizing noise above about 22500 Hz that covers whatever musical information may be there. This I believe is characteristic of the DSD recording process.

 

The remaining three recordings had spectra like the one shown in MDG_906_1363_6.JPG. There is a sharp rolloff above about 22 kHz, as with 44100 CDs. But these have rates of 88200 and 96000! More confusing still is that the liner notes indicate the MDG recording was issued as SACD. I'm guessing that these were recorded in DSD, and since DSD quantizing noise dominates above about 22500 Hz, it was filtered out in the conversion to PCM. Does anyone out there know if this is the case?

 

What worries me about 22500 Hz low pass filters is that some of them have pre-ringing. This is why Meridian introduced apodization: to get rid of the pre-ringing. I can't hear tones anywhere near 20 kHz, but I can hear the difference apodization makes on some early CDs.

 

Ray

 

 

 

[Miska]

Two recordings in my sample look like CHSA_5048.JPG. Here there is rising quantizing noise above about 22500 Hz that covers whatever musical information may be there. This I believe is characteristic of the DSD recording process.

 

Yes, this looks like DSD64 (for DSD128 the noise corner is 2x higher). The point where music's frequency content disappears in noise depends on the recording. This usually varies between 30 and 50 kHz depending on music, microphones and recording distance. However, the important thing to note is that this is different from filtering, since this doesn't cause any pre- or post-ringing. Thus, the band limitation is much more natural.

 

I'm guessing that these were recorded in DSD, and since DSD quantizing noise dominates above about 22500 Hz, it was filtered out in the conversion to PCM. Does anyone out there know if this is the case?

 

I would say that noise doesn't dominate yet at 22k, but it just starts to increase from below -120 dB, it begins to dominate higher. It is impossible to say from the spectral figures what could be the source material. That would need some hard facts.

 

What worries me about 22500 Hz low pass filters is that some of them have pre-ringing. This is why Meridian introduced apodization: to get rid of the pre-ringing. I can't hear tones anywhere near 20 kHz, but I can hear the difference apodization makes on some early CDs.

 

This is the major reason why brickwall filtering involved in CD resolution is bad. SACD (DSD64) doesn't have this hard band-limiting and in addition has better low-level linearity. For PCM too, higher the sampling rate, lower the audibility of the band-limitation. As I've said before, in my opinion the sampling rate should be high enough to cover all of music's spectral content (above converter's noise floor) without band limiting.

 

One problem I've seen is that there are various ADCs and DACs from well-known manufacturers which limit output bandwidth to 20 kHz or otherwise significantly lower than Nyquist frequency...

 

 

 

[RayW]

Miska,

 

Thanks for your comments.

 

The question remains as to why some 88.2 and 96 kHz recordings have spectra that fall off rapidly around 22 kHz. I hate to think that they have been upsampled from 44.1 kHz. If they were recorded in DSD and then converted, I wonder whether or not some DSD-to-PCM conversion software includes low-pass filtering.

 

I took a look at a 2.8 Mbps DSD live recording I made on a Korg MR-1000 and then converted to 96/24 PCM using Korg's AudioGate software. The spectrum bottomed out at about 25 kHz and then increased at higher frequencies. So this software does not low-pass filter.

 

Ray

 

 

 

[Miska]

If they were recorded in DSD and then converted, I wonder whether or not some DSD-to-PCM conversion software includes low-pass filtering.

 

Some may contain, but if there are no signs left of the characteristic noise bump, then I would suspect source being 44.1 kHz.

 

For example my software supports optional extra filters for the noise, but none of those are so steep or low corner that it would remove the noise bump completely from spectrum analysis. And the target rate is 176.4 kHz where noise is more concern than at 88.2 or 96 kHz rate.

 

 

[Shadorne]

I could not tell much from your screen shot - I am not familiar with Audacity.

 

1. Are you applying a window filter to the data set prior to FFT?

 

If not, then you need to. Analysis can be badly affected by data at both the ends of the analysis period (window).

 

 

2. High-resolution is not synonymous with high frequencies. It is very difficult to nigh impossible to judge the sound quality of a file simply based on a frequency spectrum.

 

 

 

[Mike in MD]

Hi,

I have no idea how to apply a filter to a WAV audio file. Any and all suggestions will be tried.

I agree frequency extension is not the same as sound quality, but a 24/96 file should show clean highs that go beyond 22k. Otherwise, we could simply be satisfied with 24/48. But most listeners think that 24/96 is better, and 24/192 is said to compete with good vinyl play-back.

 

Mike

 

[PeterSt]

As I said earlier (or tried to :-) : there is hardly any analogy between high resolution and high frequency (although the former is needed to let work the latter); if the file is band limited, it's no problem at all, assuming you won't perceive anything from the higher frequencies (which is a debate by itself).

 

Edit : Eh ... no problem at all goes too far because IMO it is exactly the band limiting which makes hires not sound good. But band limiting does not touch the resolution, nor the audible frequencies. Also, hires is not about high frequencies (although you would be able to calc how much can be in there), similar to HD not being about the color depth (but with SD you won't be able to effectively use the color depth HD allows).

 

That parameter you are looking for is an FFT parameter; you don't apply it to the file, but to the FFT process.

Not that I know where it sits in Audacity ...

 

 

[Miska]

2. High-resolution is not synonymous with high frequencies. It is very difficult to nigh impossible to judge the sound quality of a file simply based on a frequency spectrum.

 

Partially yes, partially no, but in many cases it is easy to detect if it's upsampled or bandlimited by using spectrum analysis.

 

1) If certain instruments don't extend above 25 kHz, there's reason to believe that the recording equipment or intermediate processing used is not up to hires standards. Recording should reproduce full frequency range instruments produce, regardless how wide it is. Challenge in itself for microphone manufacturers.

 

2) Background noise from the analog stages should be visible up close to Nyquist frequency. If it suddenly drops above 22.05 or 24 kHz there's a reason to be suspect upsampling.

 

3) If there are any remains of digital images visible above said frequencies, it is easily detectable from a symmetry in spectrogram.

 

1. Are you applying a window filter to the data set prior to FFT?

Not that I know where it sits in Audacity ...

 

It's in Edit/Preferences.../Spectrograms.

 

[wgb113]

Mike,

 

Thanks for the links and write-up. It'd be sweet if there were some sort of iTunes plug-in (a visualizer) for info like this. I'm interested to see what's doing what on certain recordings that sound off to me.

 

I also get Cookie's point 100%. I share his perspective from both ends. It's like photography...some want to have their best gear with them all of the time but what great photos do they miss out on when they don't? At the same time, I'll take a great performance over a great recording any day.

 

I'll be downloaded Audacity when I get home for pure S&Gs.

 

Bill

 

 

[Vitor]

 

Hi all,

 

I've been reading this interesting discussion and have run Audacity to see what my recently acquired HDTracks purchases were "made of".

Two sample tracks, one from the album "The World's Greatest Audiophile Vocal Recordings" by Chesky Records and the other from the recently announced "Koln Concert" by ECM.

In attachment are the screenshots, please feel free to comment.

Both tracks are supposedly 96khz/24bit.