Saturday, February 16, 2008

MP3 comparison chart of two encoders

Please notice that this is the first article i wrote abut mp3 tech analysis , depicted on 96-97 and appeared on my old geo site


UPDATES ( i think made on 91-92 ):

Following a series of upconvertions adn other experiments that itme i have found soem other peculiarities that are main to Musicmatch:

  1. Upconverting monophonic bit rates to stereo bit rates can cause stereo widening of these monoral audios. The lowest stereo mode i found or used was the 24 kbps
  2. UPconverting to higher bandwidths may cause / produce 'reaction /reactance carriers' . For example a 8 kb ( of 8 kHz) upconverted to 24 kb cana cause ahigher freq whistle to be heard, that can be soemtiems very annoying. please notic that this does not only happen between MP3 converions but also to convert one WAV file of 32 kHz to an MP3 file of 44 kHz bandwidth . There was indeed one reactance carrier of ca 16 kHz . The problem for this case has been corected with the usage of Xing encoder
  3. Some low rate ( such as 8/8 , 8/11 and 16/11 ) MP2-layer 3 files seem to use a 8 bit audio resolution. Using 8 bit resolution WAVS to converting to MP1 ( above 64 kb rtates at leat ) can cauise the MP3 files to have the half sixe!!
  4. As you can see from the frequwncy responses tables there is soem important information that shows a nrealy direct relation of bandwidth and compression And here is a over simplified relation.Notice trhat : any freqs above this band-stop are sharply removed Onfo is for Musimatch with slight improvement on Xing for the theoretical noise example
  5. 160kbit - 44 kHz gives a 20 kHz bandwidth
    128 kbit -44 kHz gives a 16 kHz bandwidth
    96 kbit - 44 kHz gives a 11.8 kHz bandwidth MPEG 1
    64 kbit - 22 kHz gives a 10.4 kHz bandwidth
    48 kbit - 22 kHz gives a 6 kHz bandwidth
    32 kbit - 22 kHz gives a 5,2 kHz bandwidth
    16 kbit -11 kHz gives a .4.7 kHz bandwidth
    8 kbit -8/11 kHz gives a 3.3 kHz bandwidth MPEG 2

  6. One very important point when using players with low bit rates. It is very important to disable Doirect X from the player otherwise audio will be strongly


Following a long period of digitizing my music collection into MP3 I have also experimented with two different popular encoders and below you will find some very important results:
The experiments are two based upon a standard method .( tones and white noise)
Program for making these is CoolEdit Pro (trial) with trial MP3 plug in (read only)

Step A Making the files

A 1:

Tones produced are : 3 secs each tone , 0 db level , with the following frequencies ( all are Hz)

  • 1000
  • 5000
  • 7000
  • 9000
  • 13000
  • 14000
  • 15000
  • 15500
  • 16000
  • 18000
  • 19000

and saved as WAV 44.000. mono , 16 bit resolution

A 2

Making white noise again under CoolEdit with:
1 sec blank
3 secs stereo white noise of 0 db level
1 sec blank

with 48000 kHz , 16 bit resolution

Step 2- compressing

Using the two compression programs ( Music match 5 and Xing encoder 2.1) step by step in various compression levels :
VBR - 160 – 128 – 112 – 96-80- 64- 56-48-32-16- 8 then save each produced file by compression – name
All files are then archived into a ziped file for later references
And here are some interesting and funny results ater zipping :
WAV files are expected to be compressed , nearly to the level of standard Xing compression
Thed tone file was 34.560.444 bytes and zipped to 558960 bytes
But still , Xing , tone compression , seems to still compressed at high rates . files of abt 720 kb have been compressed to only 9200 kb! See below for more

Step 3- checking compressed files back on Cooledit – commentary

Cooledit pro has been again used for chacking the produced MP3 files . After loading files a smoothing process has been calculated ( RMS levels )

Part 1 the tone response

Very surprisingly both encoders shown a ‘peculair’ attitude towards the pure tones by producing faint signals of lower levels , up to –8 db .
Both compressors did not respond to tones above 17000 kHz leaving only blank audio and ‘clicks’ between tone freq change

Music match 5 , shown a very big number of faint signals ( about 15 .) nearly as it happens with Xing encoder under VBR but levels were higher than of Xing . There was only the restriction of the bandwidth which nearly corresponds to the compression rate . More on this later
Xing however under 80-160 kb rates produced very clear ( faint free ) tones . IN some cases upper freqs ( below 17kHz) shown some ‘ripple’ ( on off ) faint signals of very low levels some hundred Hz above the main freq. VBR however was relatively ‘dirty’ as shown in the table below ( levels of 40 db or more

IN both encoders , tones of higher main freq. than bandwidth ( esp for rates below 48) produced mirror tones of 36 –40 db lower than main tone level

I have made also a try with WMA module of MM5 with results nearly of the MM-MP module

Part 2 the white noise response

While original signal was with exact edges between start and stop of noise, all MP compressed files shown a fade up and fade down of audio response ( either time or frequency domain ) as shown in the spectral view.

The table 3 shows the response of the 24kHz stereo white noise showing the decibel points

While Xing had the same audio level , MM5 had a 6 db loss

Xing also had nearly flat freq response at 99% of the spectrum while MM5 had a slow decrease response over higher freqs ( points shown )

Also 112- to 160 rates on Xing have the same response.. Though io could not see any significant differences in the spectrum alnalyzer of Cooledit, this can possibly lower the dynamic range .

MM in contrast shown a significant decrease of response dropping from 20 to 16 and 14 kHz for the 160 , 128 and 112 kb respectively ! Though playback on Winamp or MM player shows a 44 kHz range ….

Comparing to the response curves , in all cases except 16 Xing has higher response of 4-5 % than MM5

Special commentary on playback of the WAV tone file

Cooledit , as expected as playing very clear the tones .
However direct connection of line out from the SB 128 to the headphone showed nearly no audio on freqs above 9000 Hz ( mismatch??)
In contrast to it the HIFI connected same way produced all the tones till 19000 hz ( audible from the headphone output from HIFI )
Nullsoft’s Winamp 2.7 had a totally MP3 type audio (producing the ‘ghosts’or spur tones as shown in the Xing )
Also Media player had the same ‘attitude’
In contrast to it real Player produced clear audio in all the tones !

Add on : Time rates for compressing a music file

The table is self explanatory. Notice that this has been made last year when I used the Pentium 133 with 64 Mb memory (which possibly does not matter on the compression speed so much )
The actual WAV/ PCM file is a song of 5 mins 27.86 sec of a total 57.8 MB size
First list shows the constant rate while second list shows variable rate centered at 80 kb and variations of the shown level
The 3rd list shows times produced with the MP 3 compressor another program capable to produce MP compatible WAV files ( it possibly uses the Microsoft MP encoder) . IT also produces 8 kb/8kHz files


LIST 1

Responses of clear tones on Xing encoder on various compressrion rates


lo - nor Xing Lo var Xing Xing 160 , 128 112 96 and 80 kb 64 56 48 32 16
1000Hz 12 ……….1000

-40……….5200
-42 ……….7200
-48 ……….11600
13600 ……….1800

-38 ……….5100
……….7220
……….11500



3000 50000 7000 9000 11000 with -40 3000
5000
7000 /-47
9 000
1000/-50
3000/-36
5000
7000/-24
5000 Hz -42……….200
-32……….1200
-30 ……….2800

+12……….5000

-24……….7500
-38 ……….9600
-36……….11200
-44……….12300
-42……….14000
-48……….14600

……….200
-32……….1100
-30……….2500
-25……….7500
-37……….10000
9000,

11, 13, 5 17, 19 21 kHz with –72

9000,

11, 13, 5 kHz with –72

9000,

11, 13, 5 kHz with –72

1000 3000 7000 9000 1000 ……….-36 3000
5000
7000 /-32
9 000
1000/-40

7000 hZ (+12) -40……….200
-34……….700
-32……….2100
-44……….2600….
-22………4500…
> -36……….12000
-38……….13200`14200
-32……….2100
-24……….4500
-18……….9100
-32……….11900



Clear Clear
9000 Hz -42……….700
-38……….1800
-30……….2700
-32……….4200
-20……….7500
-13……….11300
-30……….13900
-38……….15200…
&##8230;….16100
……….500
……….1900
……….2900
……….4900
-20……….6780
-15……….11200
15 and 21 kHz with-60 db 15 with-60 db - 3000 ghost @-24 3000 ghhost/-24
13000 -38……….1000
-40……….1200
-30……….5800…
𕢦….6800
-26……….8100
-14……….10700
-9……….15400
-26……….17800
-24……….19000
……….400
…… .1000
…….15600
-36.….3500
……...3800
-32….5800
…….6200
-32….6800
-26….8200
…….8500
…….9100
-14….10600
Clear Clear Is cldear ( no signal) 10000 ghost @-32

14000 -40……….200 1200 2000 2800 48005200 10100
-32……….41200
-30……….6800
-24……….7800 .9200
> -13……….11800
-8……….16500
……….15000 .15500
only for >30

-30….7800…8100
-24….8800….9100
-24….10100.11200
-16….12600..13100
-6….17600….18000


-72 with

2000

6000

10000

18000

20000kHz

Is cldear ( no signal) Is cldear ( no signal) 9000 kHz ghost @-20
1000/-24
15000


Is cldear ( no signal) Is cldear ( no signal) 8500/-24 db ghost
5000/-24
15500


Is cldear ( no signal ) Is cldear ( no signal) 8000 /-32 7000/-23 0/-24
16000 Is cldear ( no signal ) Is cldear ( no signal) Is cldear ( no signal) Is cldear ( no signal ) Is cldear ( no signal) 7000 /-23 db 8000……….30 1000/-24
18000 Is cldear ( no signal ) Is cldear ( no signal) Is cldear ( no signal) Is cldear ( no signal) Is cldear ( no signal) 5000/-24 db 5000 ghost /-24 2000/-24
19000 Is cldear ( no signal ) Is cldear ( no signal) Is cldear ( no signal) Is cldear ( no signal) Is cldear ( no signal) 4000/-24 db
3000/-24

Lo - nor Xing Lo var Xing Xing 160 , 128 112 96 and 80 kb 64 56 48 32 16


FILE SIZES

Rate

Xing Size

Xing Ziped

16

71928

5159

32

143904

8748

48

215856

14533!

56

252158

7722

64

288192

8263

96

432288

8151

112

504336

8653

128

576384

8947

160

720480

9227

LO NOR

287296

216502

LO

336871

274009



RESPONSES ON WHITE NOISE FOR MM5 AND XING ENCODERS

WAV FILE HAS
1 SEC BLANK – 3 SECS WHITE NOISE – 1 SEC BLANK

AUDIO LEVEL IS +3 DB

Rate

xing encoder

M M 5

FILE SIZE XING

var low -6 db<13.5khz
30864
var lo no -6 db -60 db 20
36027
var no -6 db -60 <20>
42860
160 0db -38db/21. -2/2k
-6/16k
-11/ 20k
100320
128 Same 2 to 6 up to 16 80256
112 Same -6 to 14k 70224
96 0db <15....> -6 /11.8 60192
80
0 <10.1>
constant to 11000 @-6 50400
64 0<3,4> const to 10.4@-6 40320
56 -6 /9.2
-45/9.6
-6 to 8.8
-60 /9.2
35280
48 0
-2 /3.4
-6/7.6
-6/6k
-60/6.2
30096
32 0
-2/5.1
-60/5.5
-6 /5.2
-60/5.3
20160
20
-6 /3.8
-60/3.9

16 -6/4.0…
-60/4.1
-6 /4.7
-60/4.8
10008
8
-6/3.2
-60-/3.3


TABLE 3

Times compression ratios on A FILE

Using A 133 Mhz Pentium 1

file : wav 44 khz 16 bit size 57.834 kB 5:27.86 sec (327.86 sec)

Music Match 5 under

Mode Kb Range Fielsize Comp rate Time elapsed
8 8kb 11khz 328 kB 1/176 1:02 min
16 16 11 656 88 1:19
18 18 11 - - 1:22
64 64 22 2623 22 2.22
96 96 44 ~3924 15 2:45
128 128 44 5246 11 2.51

~ estimated

v 1%
78-79 3222 18 2.49
15%
81 3366 17.1 2.46
88%

5982 9.6 3.00

mp3 compressor

16 16
657 89 1.43
128 128 44 -
>5 min

No comments:

Post a Comment