Posible bug in built-in MP3 decoder - please verify

[sikahr]

Procedure:

1. I generate 1KHz sinus -90db test signal with wavosaur 1.0.0.9000.
File: "sinus 1kHz -90dB.wav"
Format 24-bit mono wav
Playing with SQ3: OK

2. I encode "sinus 1kHz -90dB.wav" to mp3 with lame 3.97
command: lame --preset insane "sinus 1kHz -90dB.wav" "sinus 1kHz -90dB.wav".mp3
resulting File: "sinus 1kHz -90dB.wav.mp3"
Format MP3
Playing with SQ3: strange arthefacts in sound louder then test signal, very weird

3. Now I decide to decode MP3 on PC and play decoded wav file on SQ3.
I decode "sinus 1kHz -90dB.wav.mp3" on PC with 2 mp3 decoders:

A. Lame
command: lame --decode "sinus 1kHz -90dB.wav.lame.mp3" "sinus 1kHz -90dB.wav.lame.mp3".lame16.wav
resulting File: "sinus 1kHz -90dB.wav.lame.mp3.lame16.wav"
Format 16-bit mono wav
Playing with SQ3: Distorted 1Khz (no dithering), OK for decoder without 24 to 16 bit dithering

B. Mad 16 bit
command: >madplay -v -b16 -owav:"sinus 1kHz -90dB.wav.lame.mp3".mad16.wav "sinus1kHz -90dB.wav.lame.mp3"
resulting File: "sinus 1kHz -90dB.wav.lame.mp3.mad16.wav"
Format 16-bit mono wav
Playing with SQ3: Clean 1Khz with dithering noise, OK for decoder with 24 to 16 bit dithering

C. Mad 24 bit
command: >madplay -v -b24 -owav:"sinus 1kHz -90dB.wav.lame.mp3".mad24.wav "sinus1kHz -90dB.wav.lame.mp3"
resulting File: "sinus 1kHz -90dB.wav.lame.mp3.mad24.wav"
Format 24-bit mono wav
Playing with SQ3: Clean 1Khz without noise just like original file, OK for full 24-bit decoder

Additionaly, I normalized wavs to -6dB to hear diferences without extreme amplification:
Normalized -6dB wavs:
"sinus 1kHz -90dB.nor -6dB.wav"
"sinus 1kHz -90dB.wav.lame.mp3.lame16.nor -6dB.wav"
"sinus 1kHz -90dB.wav.lame.mp3.mad16.nor -6db.wav"
"sinus 1kHz -90dB.wav.lame.mp3.mad24.nor -6dB.wav"
Of course, to hear how SQ3 play problematic mp3 you must listen SQ3 directly with extreme amplification.

4. Conclusion
Built-in MP3 decoder in SQ3 is problematic/flawed with some files, PC MP3 decoders (MAD and Lame) decodes that files correctly.
Polite request to Sean Adams/Slimdevices: please fix it, best to implement full 24-bit MAD decoder
as stated on Squeezebox product info web page: http://www.slimdevices.com/pi_squeezebox.html
# MPEG decoding uses MAD software, widely regarded as the most accurate, most compatible MP3 decoder.
# High accuracy 24-bit synthesis
# Supports all MP3 data rates and sample rates, including VBR


5. Someone please test this procedure on your system, to verify my facts.
If someone have sound card with s/pdif in please capture output from SQ3 when playing problematic mp3 file
("sinus 1kHz -90dB.wav.mp3") for further analysis.



Remarks
-my system: windows Xp SP2, slimserver 6.5.3
-all wavs are 44.1KHz
-all files except normalized are extremly quite, to hear something you must amplify to the max,
I use SQ3 on 100% volume + headphone Amp with 11x amplification on Sennheiser HD580
please be very carefull when work with maximum volume beacuse you can easily damage your ears/equipment,
start with low volume, then gradualy amplify
-Used free software links:
Wavosaur free audio editor http://www.wavosaur.com/download.php
LAME 3.97 release http://www.rarewares.org/dancer/danc...f=lame-current
MADPlay v0.15.2b Command Line mp3 Player/Decoder http://www.rarewares.org/files/mp3/madplay-0.15.2b.zip

All files mentioned here are available in zip file on link:
http://www.mytempdir.com/2064518
contens of zip file:
Directory of c:\slim\zip\t

20.11.2007 14:19 <DIR> .
20.11.2007 14:19 <DIR> ..
20.11.2007 14:00 3.469 description.txt
03.10.2006 14:33 520.192 lame.exe
20.11.2007 09:02 31 lame16.cmd
20.11.2007 09:01 35 lameencode.cmd
20.11.2007 09:09 39 mad16.cmd
20.11.2007 09:10 39 mad24.cmd
23.08.2004 10:09 137.728 madplay.exe
20.11.2007 14:19 <DIR> normalized
20.11.2007 08:35 3.969.564 sinus 1kHz -90dB.wav
20.11.2007 09:01 1.202.676 sinus 1kHz -90dB.wav.lame.mp3
20.11.2007 12:04 2.647.434 sinus 1kHz -90dB.wav.lame.mp3.lame16.wav
20.11.2007 09:09 2.649.644 sinus 1kHz -90dB.wav.lame.mp3.mad16.wav
20.11.2007 09:11 3.974.444 sinus 1kHz -90dB.wav.lame.mp3.mad24.wav
20.11.2007 14:19 0 t
20.11.2007 09:00 311 Urls.txt
30.06.2007 23:49 458.752 Wavosaur.1.0.0.9000.exe
15 File(s) 15.564.358 bytes

Directory of c:\slim\zip\t\normalized

20.11.2007 14:19 <DIR> .
20.11.2007 14:19 <DIR> ..
20.11.2007 13:43 3.969.564 sinus 1kHz -90dB.nor -6dB.wav
20.11.2007 10:27 2.647.954 sinus 1kHz -90dB.wav.lame.mp3.lame16.nor -6dB.wav
20.11.2007 10:25 2.650.164 sinus 1kHz -90dB.wav.lame.mp3.mad16.nor -6db.wav
20.11.2007 10:28 3.975.050 sinus 1kHz -90dB.wav.lame.mp3.mad24.nor -6dB.wav

[sikahr]

less "technical" description of problem:

There is a mp3 with very very low volume.

When playing this mp3 directly with squuezebox there is a loud strange noise (loud means louder then signal, not absolutely).

When first decoding this mp3 on PC to wav and then playing that wav on squuezebox sound is more or less OK.


My conclusion: inbuilt mp3 decoder is in a way broken.

[peter]

sikahr wrote:
> less "technical" description of problem:
>
> There is a mp3 with very very low volume.
>
> When playing this mp3 directly with squuezebox there is a loud strange
> noise (loud means louder then signal, not absolutely).
>
> When first decoding this mp3 on PC to wav and then playing that wav on
> squuezebox sound is more or less OK.
>
>
> My conclusion: inbuilt mp3 decoder is in a way broken.
>

Alternative conclusion:

Your MP3 file is in a way broken but different decoders react
differently to it.

These things are fairly common in IT.

Regards,
Peter

[sikahr]

Maybe, but i don't believe. I generate file with lame, no errors, on PC decoding is without problem, only one problem is decoding on squeeze. Additionally, I can generate similar file (I tried sinus 1kHz -100 db), problems haven't gone away.

Regards, Nenad

[Ben Sandee]

On Nov 20, 2007 2:12 PM, sikahr
<sikahr.30d5az1195589701 (AT) no-mx (DOT) forums.slimdevices.com> wrote:
>
>
> My conclusion: inbuilt mp3 decoder is in a way broken.

It sounds to me like you've done a fair amount of analysis -- nice
work. I think you should file a bug with all the appropriate attached
data at bugs.slimdevices.com. If it's a decoder problem in the
firmware then only a Logitech employee can really contribute, since
the firmware is closed-source.

Ben

[funkstar]

why is this a problem though?

Sure it is a bug, but why were you testing for it? To me, it sounds a little like the Excel bug that appeared a few months back, here some forumulas resulting in 65,535 would appear as 100,000.

http://www.joelonsoftware.com/items/2007/09/26b.html

Yes it is a bug, but it only happens in 12 out of 18,446,744,073,709,551,616 possible floating point binary numbers. So really it isn't as sever are you would think.

[snarlydwarf]

why is this a problem though?

Sure it is a bug, but why were you testing for it? To me, it sounds a little like the Excel bug that appeared a few months back, here some forumulas resulting in 65,535 would appear as 100,000.

Actually I think it is less important than that.

Is it a bug that when you crank up the volume on a really low volume mp3 you can hear artifacts?

It seems like expected behavior to me.

[sikahr]

Please try to listen, you will get clearer picture this way.

sin 1KHz -80 db also have artifacts.

[sikahr]

It sounds to me like you've done a fair amount of analysis -- nice
work. I think you should file a bug with all the appropriate attached
data at bugs.slimdevices.com. If it's a decoder problem in the
firmware then only a Logitech employee can really contribute, since
the firmware is closed-source.

Ben

I already contacted slimdevices support.

Someone with soundcard with spdif in and squeezebox please capture digital output from squeeze when playing problematic mp3 and post it somewhere. Thank You.

[snarlydwarf]

A. Lame
command: lame --decode "sinus 1kHz -90dB.wav.lame.mp3" "sinus 1kHz -90dB.wav.lame.mp3".lame16.wav
resulting File: "sinus 1kHz -90dB.wav.lame.mp3.lame16.wav"
Format 16-bit mono wav
Playing with SQ3: Distorted 1Khz (no dithering), OK for decoder without 24 to 16 bit dithering

If I understand that right, it is the following:

1) 1Khz sine wave at -90db.
2) Converted with lame to an mp3.
3) Decoded with lame back to a wav.

Played and it had distortion on the SB3, but not on your PC.

That points to a flaw in your methodology and contradicts your claim: there was no use of the SB3 firmware for decoding that mp3... you used LAME. Yet the SB3 is 'distorted' and the PC isn't?

Perhaps your PC sound card is resampling to 48khz and confusing the issue.