Mp3 War II

I started to write my first impressions of mp3 encoders 2 years before (my first Mp3war article published on July '98) . Since then I read so much reports that I wanted to make a new review . Although I made the same hard job, I do not have the courage to make a so descriptive report . So pls accept my results in form of faq .

Q: Which Encoders did you test ?.

A:
BladeEnc 0.91
BladeEnc 0.91 dll
Gogo 2.26a
LameDLL 3.65
Lame 3.65 (w -h -f -k -m s)
ScMpx 1.51 (st / jst)
-
Xing mp3 Encoder 1.5
-
FHG dll 1.263 (w f)
Fhg mp3 Producer Pro 2.1 ( norm / hq)
AudioActive 1.54 (w -h)
-
96K - SoundVQ 2.54b6 (hq / norm)
80K - SoundVQ 2.54b6 (hq / norm)
-
128K - MsAudio4 (v.2)
096K - MsAudio4 (v.2)

I. Mp3 Tests

Q: Whow did you make your tests ?. How can I reproduce them ?.

A:
My tests are based on samples & methods I used for my first tests . There are some songs out of my entire collection that had the most audio artifacts (read Mp3War I for more info) . The resulted mp3 files played both on MS Player 6.4 & Winamp 2.6 . I tested only the critical 128K bitrate . Most mp3 weaknesses - as described in my previous article - appears in analog recorded - vinyl - old songs . 2 of my 3 sample WAV are vinyl recorded songs . New productions & techno - pop songs works OK with most cases .

 

Q: Which is the best mp3 encoder no matter process time ? .

A:
1 - Gogo 2.26a (-m s)
2 - Xing mp3 v.1.5 (-s)
3 - BladeEnc 0.91

Gogo hit #1 . Xing Mp3 lacks a bit in snares and other hi-freq instruments volume , even with hi-cut filter disabled . BladeEnc lacks in vocal frequencies were it creates something like phase effect . Gogo also creates such effect but is less noticeable . Although most programmers choose Joint Stereo for this bitrate , I found that (Simple) Stereo is most appropriate for these "difficult" samples and produces less phase effect . However I agree than in most of times (specially in modern pop/dance music) is better the default (Joint Stereo) setup in order to cover more dynamic range .

Some brief description about the other - out of contest - encoders . The so popular, so expensive, so 'pro' Fraunhoffer encoder ( mp3 Producer Pro , Audio Active Studio , FHG .dll ) was producing strong phase distortion , occurred to 8 out of 10 vinyl recorded samples !. If you don't believe me make yourself a similar test: record from vinyl any song you like that have many & continuous snares . 

The great question is that I found Gogo far better than LAME , (since Gogo is based on a prior older Lame version) . In most of my samples Lame was producing stronger trebles phase effect from Gogo , Xing , BladeEnc no matter what parameters I gave on it . I know that Lame is developing and becoming better every day , I know that most sites rate it as #1 , but I know how MY samples are heard , and my samples are my songs . As I said to my firsts tests: Don't trust your eyes , trust your ears  !!.

 

Q: Which is the best mp3 encoder counting also process time ? .

Mp3 Power Points: 
- Two of the best mp3 encoders today - Gogo & BladeEnc - are FREE!
- Two of the best mp3 encoders today - Gogo & BladeEnc - are multi-platfom (Win / Linux / BeOS etc)
- The expensive standard FHG Mp3 Encoder has a lot lower quality .
- If you compare the previous results 2 years before , ( BladeEnc was #1 in quality , Xing was #1 in speed ) nothing changed so much .
- Viva la revolution !.

II. Mp3 w VBR

VBR (Variable Bitrate Rate) was a later mp3 format option introduced by Xing at August 1998 . It permits to encode not at a fixed rate depending on the sample .

Generally speaking files created with VBR sound like one scale higher . For example if we make 1 min wav to 1MB mp3 w VBR sounds "like" 160Kbps mp3 which is 1.2 MB . Remember that "sounds like" is not "sounds as" .

Note: 
- VBR is great but be sure if your player(s) support it .

Gold Tip:
- No matter which encoder I was using , in most cases I found that:
- 128K sized VBR (or less) sounds better than 128K CBR .
- 160K sized VBR (or great) sounds worst than 160K CBR .
- So if you are willing to sacrifice more than 128K file space , return to CBR .

- Typical Examples for 128k sized mp3 :
Gogo 2.26 : gogo.exe t1.wav -br 56000 -v 7
Xing 1.5 : x3enc.exe t1.wav t1.mp3 -v50 -s

Typical Examples for CD-Rip using CDFS drivers :
- Gogo 2.26 :
for %%f in (f:\stereo\16bit\44100hz\*.wav) do gogo %%f -br 56000 -v 7 -o c:\demo
- Xing 1.5 : 
x3enc f:\stereo\16bit\44100hz\ c:\demo\ -v50 -s

Xing produces more 'predictable' file sizes , i.e. when we use -v50 parameters the files matches very closely to an 128kbps mp3 file . However Gogo (with -br 56000 -v 7) varies the size between 110 - 128Kbps . The -br command set the minimum bitrate so it's not critical . The -v 7 command is mainly responsible for it's size . If we replace it with -v 6 the size will match like an 128 - 160 Kbps file .

Q: Which is the best mp3 encoder w VBR support ? .

A:
1 - Gogo 2.26a (-m s -br 112000 -v 7)
2 - Xing mp3 v.1.5 ( -v50 -s)
3 - Lame 3.6

Very few encoders today supports VBR (Xing, Lame, Gogo) . Again here Xing lacks a bit in snares volume , Lame creates noticeable phase effect .

Q: Which is the best mp3 encoder counting VBR support and also process time? .

A:
1 - Xing mp3 v.1.5 (20" for 1 min. Wav)
2 - Gogo 2.26a (55" for 1 min. Wav)

Although Gogo was the fastest in fixed bitrate , when encoding in VBR mode it needs three times more time . When we enable VBR options , quality differences between encoders are few, because we are moving to 160Kb - sound like .mp3 . This time is measured on K6-2 / 400 and as I mentioned above the difference is greater in other CPUs . Xing seems to have no great time differncies between VBR vs CBR .

Q: I don't know much about mp3 . How can I use these encoders for CD-Rip / Encoding and use their VBR options ?.

A:
1 - If you installed the Cdfs driver , just run the batch files mentioned above . Just change drive letters & paths .

2 - Probably you will feel comfortable with "Audio Catalyst 2.x ". You don't have to install Xing Mp3 encoder (a similar engine is included in AC) . For further info read my very last question.

3 - For Gogo you can:

- For CD-Rip: Use CD-Copy or CD2wav32 (read my Cd-Rip page) .
- For Line-In (live): Use Line Ripen 1.x utilizes some encoders like Gogo . You need a very fast machine for realtime on-the-fly VBR encoding ( PII/266 or K6-2 450) otherwise use fixed BR or prefer to use Xing encoder for such job (does not need so much CPU resources) . Line Ripen can also be used for on-the-fly Cd-Rip if we installed the Cdfs drivers . You can find all URL's for all these programs at my main page .

Q: In your previous article (Mp3 War I) you recommended not to try on-the-fly encoding . Now why do you mention it ?.

A: That article was 2 years before !. Now CPU & CD-Rom got 2 or 3 times faster . You can use on-the-fly encoding if you can reach 3x or better , otherwise your CD-Rom will suffer and die young !. This method is ideal if you have slow (but good) Cd-Rom on fast PC's : It takes the same time to make either wav or mp3 , so you earn the whole encoding time . My Laptop PC has a Cd-Rom that extracts audio at 4x . So it takes the same time to make .wav or .mp3 .

II. Mp3 vs Wma vs Vqf

Q: Are there other audio formats than mp3 / wma / vqf ?

A: Yes , the major are Liquid Audio , PAC , AAC , TAC & Real Audio . I have some info that all except RA produces fine sound . I 'll test them some other time . Most of them are not practical because they are under development or they have very big encoding times (Astrid/Quartex AAC encoders needs 20 min. to encode 1 min. in a P/200 !! . If you will use Xing Mp3 it 'll take you 25-30 sec.) .

Q: Which format sounds better ? (Plus or Minus)

1 - Mp3 w VBR is the best . From the above you can understand that one mp3 w vbr file that can fit like a standard 128Kb mp3 sounds like a 160Kb mp3 . 

2 - Yamaha's Vqf is the second . The 96K .vqf (it cannot go higher) beats the rest two formats , althought I found that the very high scale of VQF (48Kb / quality) is not enough for real Hi-Fi . The distortion effect created on this format is a trebles attenuation , like on old magnetic tape players .

- Plus: Smaller files (20%)
- Minus: 
  i) A LOT of encoding time . 1 minute .wav needs 20 sec to be encoded by mp3 Gogo but 3:05 min. by SoundVQ !
  ii) Only 1 encoder (in beta time limited stage) and 2-3 players available . All in Win9x platform .

2 - Microsoft's .wma (or .asf) comes after. I found that it produces better sound than .mp3 . The distortion effect created on this format is a small echo-delay together with phase mostly in higher frequencies .

- Plus: 
1) Not confusion what to choose & where to find it ... Just go to Microsoft
2) Encoding time is good . 1 min wav takes 29 sec.
3) Microsoft gives many free applications to help you . Apart from the standard audio / video encoder you can make real-time encoding or broadcasting & others .
- Minus: No other platforms / devices available .

3 - mp3 with fixed BR (CBR) today comes last . The distortion effect created on this format as described on my previous article is a phase difference & trebles periodic attenuation . The strong advantages that made it popular remains unique :

- Public domain format . No loyalties from anyone for the use of .mp3 files .
- Multi-platform .

Q: Which format do you use ?

A: mp3 of course , but I encode @ 160Kb . I have 20% more size and I do not want to loose mp3 CBR compatibility . There are still some songs that cannot be encoded well @ 128K VBR and I'm tired of making tests and re-encoding just to gain the difference from 160K to 128K Vbr . I want to play my files on Dos mp3 players or on Rio-like devices . I don't use VBR for compatibility and simplicity reasons .

Q: Is it better an 96K .vqf or 128K .wma than an 160 K .mp3 ?.

A: just NO !.

Q: All these are too complex for me . Is there any way to rip / encode quick & simply ?.

A: Yes . Choose 'Audio Catalyst 2.x' (It's not free) . It is one of the best CD-Ripper including Xing's engine, which is the 2nd better encoder today . Supports VBR , can rip on-the-fly , can also encode wav files , supports CDDB etc . The GUI & menus are very friendly & nice . Audio Catalyst is my recommended software for on-the-fly encoding (although many times crashed my system) . It's the only CD-Rip software today capable of produce VBR mp3's in realtime . It reach 3x - 4x on my PC's . My old P/233 MMX hits 2.5x , that means that a 75 min. Audio-CD will be encoded to mp3 in 30 min.

Conclusions: 

1) The mp3 technology changed (VBR) , the encoders NOT !. Xing is still at the top .
2) Mp3 today has so much opponents than never before . Finally .mp3 did not defeated by the big companies like Yamaha's .Vqf or Microsoft's Audio 4 . But even if it will do , the point is that people are not willing to change it for something better if it's not free and widely accepted .

30/03/2000

P.S. MusicMatch Jukebox 5.x today makes also fast on-the-fly .mp3 and uses the new FHG 'FastEnc' . Although it claims that it's the worlds most hi quality encoder and I read many articles putting it #1, I did not found many differences with previous versions , it's still fails to encode well on my samples and it's still out of contest on my tests .

10/07/2000

P.S.2: Also until now I read many reports saying that LAME is #1 (see my links on main page) . So I thought that something was wrong and I tried again & again with new samples & playing with LAME parameters (my last LAME ver. was 3.87mmx) . My resulted opinion did not change: Don't trust your eyes , trust your ears !. LAME produces most noticeable phase effect . Probably Gogo looses trebles more easily that's why it looses points , but a phase effect is more disturbing than a (probable) treble loss , isn't it ? .

01/11/2000

Regards
Michael Hondrosmos