Signals that Trip-up ATRAC

SQAM Signals

A set of EBU SQAM recordings ("Sound Quality Assessment Material") represent particularly difficult audio signals for perceptual coders (you can torture-test your MD with these).

Sony and Sharp ``Warbling''

The Titanic soundtrack, Track 7, at around the 34 second mark causes ATRAC to falter. My cousin first noticed this while listening in a very quiet environment, on a disc recorded on a Denon DMP-R70, which uses Sharp ATRAC 4. The same track, recorded on my JE510 (Sony ATRAC 4), is far superior but the artifact is still noticeable if you really listen hard.

So what is the artifact, I hear you ask?

It's a high frequency sound that has a warbling effect. In the original, the sound is there, but is very smooth. I'd like to hear what ATRAC 1 would do to this, or MP3 at 128Kbps - it would be quite severe no doubt.

-Colin Burchall ([email protected])


Sharp ``French Horn'' problem

This short section of a French Horn solo, when recorded on a Sharp 722, is played back with a strange spike.

-Colin Burchall ([email protected])

Mathias reported this problem as well.

Tim H. reports:

I've played with the wave files. I own a Sharp MD-MT15 with Sharp ATRAC 6, and can reproduce the problem by recording in analog. Don't have a digital out, so can't test it in digital. (It would help if the file was standard 44.1KHz, too)

  1. If you look at the last "click" in the 722 wav file, you can clearly see that the waveform looks like it's been clipped and pasted, ie, they retain the shape they're supposed to, but doesn't line up with the rest of the wave form, and thus make that 'click'. It's the same for the other clicks.
  2. If you swap the left and right channels and record them, the clicks stay in the right channel.

So, I think there is a bug somewhere either in the ADC or the ATRAC calculation in the right channel. It would be nice to record the original wave form in digital (and 44KHz) and rule out the ADC. I doubt Sony units do this, because it really looks like a simple bug, rather than some error in the Sharp psycho-acoustic model.

Christian Kaunzinger reports:

Maybe the reason for the ATRAC-breakdown, or whatever it is, is not the recording but the playback section of the Sharp MD-recorder.

I experienced the same error on some of the recordings on a Sharp 701. The sound that appears at some points sounds a bit like a clipping signal caused by digital overdrive. The recordings were made with different CD-Players and on MD's of different brands. The errors only occured when recording digital, I never experienced any errors when recording with analog cable.

To find out about the reason for the errors I made a short test with two MD recorders, Sharp MD-MS701H(S)2 and Aiwa XM-R70. A track (see below) was played with the CD player (Technics SL-PG490)and recorded with the digital cable on the recorders.

Result:

  1. Errors could only be heard when playing the disc on the Sharp. When playing both recordings on the Aiwa, no drop-outs could be found !

  2. Playing both the Aiwa and Sharp recorded tracks on the Sharp shows that most of the errors are at the same points. However, a few appear only on the Aiwa-recorded track and a few others only on the Sharp-recorded track. Some of them are very loud and others can be hardly noticed, especially when they go with a high snare stroke.

  3. The error does not appear by chance, e.g. when shaking the recorder, but always when the track is played.
Finally- the reason may not be the recording, but the playback section of the Sharp.

CD used:

The track that causes lots of those clipping like signals can be found on the CD 'Red Hot & Cool'. It's the 5th track called 'Flying high in the brooklyn sky'. times: 00:37, 00.59, 01:40, 1:30, 01:50, 02:38, 02:48, 04:00, 04:39, ...

Equipment:

CD-Player: Technics SL-PG490, Technics SL-PG590, Sony CDP-XE530
MD-Recorder:Sharp MD-MS701H(S)2, Aiwa XM-R70
All recordings were made with digital cable.

ChrisO reports:

I don't have any rigidly-controlled and repeatable examples of how to bust the ATRAC compression algorithm, but I do have some experience with portable MD equipment creating the exact type of waveform distortion that is illustrated above.

I was recording a folk-rock band at a club using a Sony MZ-R50 and a pair of AT cardiod mikes. My levels were set fairly conservatively (I may have gone over once or twice during a 90 minute performance), and neither the recorder or the mikes were disturbed during the show, except to move the deck once to change discs.

When I played the discs back in my car on the way to my hotel after the show, I was disturbed to find that the recording was peppered with very prominent clicks. I later digitized the show onto my computer. When I looked at these clicks in my sound editor, I found the exact distortion pattern pictured in the image referenced above hundreds of times throughout the recording.

This distortion pattern was found in both of the discs I used that night (2 different brands), occurred roughly equally in both stereo channels (but never in both channels at the same time), and was loosely correlated with the sound of a loud kick drum. (I say "loosely" as the distortion was often introduced when the kick drum was struck, but it also happened at very quiet times when the drummer wasn't even behind his kit!) After playing with the errors in my sound editor for a while, I realized that this distortion was basically an inversion of the sample values that should have been recorded, with the exception of the "ramp" samples at the extreme edges of the click where a sample or three were recorded between the proper sample value and the inverse of the proper sample value.

Of course, it was not truly an inversion (which would have been easy to fix in my editor...) --- the sections of the signal which appeared inverted exhibited the same "wobbly" pattern that is shown in the example graphic instead of a clean inverse of the curve that would have been there if this were a true inverse of the signal.

I used the same equipment a few nights later without further incident.

I'd absolutely love to find out what caused this to occur. But I think that is somewhat of a tall order at this point. So please accept this as anecdotal information.

Adam F reports:
I tried that horn sound out on recording to my Sharp MD-MT831, i experienced no such artifact. However I tried it out with 722, and I got the artifact. I suspected that Sharp had changed a significant amount of hardware between the two portables, so I guess this would confirm it.

ATRAC ``Techno'' problem

Jake Hamby wrote:
[Perceptual encoder artifacts are likely to be] especially apparent with acoustic instruments like guitar and cymbals (which can have very complicated upper harmonics that the brain has an entire listening history to compare to), and with techno music, which often has very pronounced percussion and strange electronic effects which can be very difficult to encode (try encoding the first 30 seconds of "Kalifornia" by Fat Boy Slim with any MP3 or MD encoder, listen to the result with some good headphones, and you'll hear what I mean!).
Ian Horsey tried this and reports:
The result is that an additional high(ish) pitch scratching sound appears on the recording. It is only noticeable with headphones though. The tests I did were made by recording digitally onto a Sharp MD-MS702 and a Sony MDS-JE520, and both produced identical results.
Eric Walli confirms:

Source: Technics SL-B101 turntable w/Grado ZT+ cartridge
Receiver: Marantz SR5100
Recorder/Media: Sony MZ-R90/Memorex 74
Album: Fatboy Slim - "You've Come a Long Way, Baby"
Track: "Kalifornia" (Side two, track three - played ~5 times)
Speakers: Advent 12XP (2), JBL 500 (2)
Headphones: Koss R/10

I set the recording level to peak just below the "over" indicator. Headphones make the artifacts easier to hear, but they're still audible on my Advent speakers. The first 16 seconds (nothing but vocoder) sound as though it was recorded from a badly worn record; there's a lot of scraping and scratching. Once the drum machine comes in, however, the scraping gets masked and becomes incredibly hard to detect, even through headphones.

While it was recording, I monitored the signal coming off the record. The scratchy sound wasn't there, so I can only guess that the ATRAC encoder was stumbling over the wacky sounds coming from the vocoder. While everything was hooked up, I switched the source from the turntable to the computer and recorded the French horn sound on the "Signals that trip up ATRAC" page. I used a 1995 Ensoniq Soundscape Elite soundcard. The only artifact occurred at the very start of the file, and the click was very similar to the sound the Sharp made.

Much like other compression formats, ATRAC tends to have trouble with dance and techno - Depeche Mode's "Enjoy the Silence" from a CD has the same "worn out record" artifacts as mentioned above, especially on the bass drum kicks.

MP3 is noted for horrible sound, particularly on Collective Soul's "Shine" and "December". TwinVQ/VQF's major artifact is an overall "deader" sound, partly due to a blending of high frequencies between the left and right channels -- however, at 96 kbit/sec, it surpasses MP3 at 192 kbit/sec. A comparison between the two formats with Radiohead's "Fake Plastic Trees" with a constant crash cymbal ride towards the end shows the difference well, though neither of these formats are quite as good as ATRAC, IMO.

All observations here should only be taken as opinion. Like so many other things in the world of technology, YMMV.


Sharp hiss on Cello solo

I'm still looking for that piece of music that will defeat the ATRAC on either my Sony R55, Aiwa F70 or Denon DMD1000, but I did manage to defeat the ATRAC on the Sharp 821 purely by accident! I was using a Marantz CD-48 CD player (optical output) recording onto the Sharp MD-MT821 using the digital input. Recording levels weren't adjusted at all.

I used the Deutche Grammophon (Polygram) recording of Dvorak Cello Concerto in B minor (CD ref: 439 751-2) and right at the start you can hear a definate "hiss" like steam every time the cello goes near the low frequencies. This is only really noticable in the solo passages, although it can be heard through the orchestra in some of the quieter parts too. The sound was not there when I tried re-recording with either the Aiwa, Sony or Denon unit, leading me to believe it is the Sharp ATRAC system.

I would be much obliged if somebody with access to this CD and an MT821 could try to reproduce this, just in case it is the unit I was using. I tried a TDK, Maxell and MMore disc, and the results were the same.

-Matt Charman


Warble on Pioneer MJ17D

The AC/DC song "Highway To Hell" causes the Pioneer Elite deck to throw up. As soon as the high hat starts in you can definitely hear a high midrange/lower treble 'warble' affect that is not present on the CD. The Sony MDS-JA20ES deck and the MDS-JB920 do not do this, they are clean, but the Pioneer really does have a problem with it.

-Gerry Masters


Sharp Peak Distortion

The nylon-string(?) guitar opening to Edie Brickell's "Circle" from the "Shooting Rubberbands at the Stars" album, when recorded (and played back) on the Sharp 702mk will sound as if it has been overrecorded-- it's "smeared"-sounding-- no matter whether recorded digital or analog, even if the analog level is reduced. The effect is most noticeable on the loudest peaks, which adds to the mistaken impression of overdriven record levels (as in an analog tape system).

The Aiwa AM-F70 does better, and so does the JE530 deck. None is perfect.

-Walt Brand


Sharp Frequency Warble

I have a Sharp 831 and I find that pretty much ANY signal that contains a reasonable spread of frequencies (as long as there are prominent high-end frequencies) ends up getting encoded to sound 'warbly'. Specifically, cymbals, or crazily-distorted guitars

Example tracks would therefore be:

-Dave Hooper


Sony trouble on Piano Transients and Hi Hats

CD 'At the Drop of Another Hat' - Parlophone CDP 7974662. I can hear clicks on the MD dub. The worst offender is track 8, 'Friendly Duet' from about zero to twenty seconds. The only clue I can find is that these are hard piano tranients. Curiously, when the vocal starts the clicks vanish (piano record level reduced?)

I tried three transfers: first from a Sony CDP-M77 player via an analogue link to a Sony MDS 501 (ATRAC 2); results bad. Second attempt was on a Sony MXD-D1 (internal - digital link, presumably) (ATRAC 3, I believe). Just as bad. Third attempt, recording from the CDP-M77, via digital link to a Sony MDS-JB930 (ATRAC R). Better, but still clicky. Reducing the record level on the MD seems not to have any effect on the clicks.

Second problem CD is 'Pure Moods' - Virgin VTCD 7243 39659 2 2, track 16 'Songbird' (Kenny G). From about eighteen seconds onwards (where the sax comes in) there's a 'roughness' or 'scratchiness', particularly on the right channel. Only clue I have is the right channel has most of the hi hat information.

I used the first and third methods of transfer. The 930 showed only a small improvement over the 501. Loudspeakers were JPW self-powered monitors and Quad electrostatics with Quad amps. The Quads didn't like the Kenny G piece much at all.

-Ian Oliver


Please feel free to mail in a contribution. Don't forget to document how to recreate the problem, including the CD name, track number and time points.

Return to the MiniDisc Community Page.