Skip to main content

Notice

Please note that most of the software linked on this forum is likely to be safe to use. If you are unsure, feel free to ask in the relevant topics, or send a private message to an administrator or moderator. To help curb the problems of false positives, or in the event that you do find actual malware, you can contribute through the article linked here.
Topic: Lame 3.99.5 outputs smaller files than 3.97 ? (Read 8180 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

Lame 3.99.5 outputs smaller files than 3.97 ?

I ripped most of my CD collection about six years ago with LAME 3.97b with the V2 preset. I recently had a hard drive crash and had to re-rip pretty much everything. Fortunately, i still had all the physical discs stored and in excellent condition, so I lucked out this time. I downloaded the latest version of LAME 3.99.5 to re-rip.

As it turns out, a few albums of the original 3.97 rips survived the crash (saved to an external SD card), so I was able to compare a number of tracks. I noticed that pretty much every newly ripped song had a lower bitrate (5-10kpbs) and file size than the ones ripped with 3.97. Is this to be expected due to the improvements in compression techniques made to LAME over the past several years? I suppose I would have expected the V2 preset to hover around the same average bitrate as the encoder improves, and that it would use the extra headroom to improve quality in difficult passages...


Lame 3.99.5 outputs smaller files than 3.97 ?

Reply #2
Remember that the goal of VBR lossy compression is to achieve a target quality at the lowest bitrate possible, so you are witnessing the improvements made by the LAME developers since 3.97. The new VBR mode is also much faster than the old one in 3.97, so continue your encoding process and enjoy the smaller files and the saved time.


Lame 3.99.5 outputs smaller files than 3.97 ?

Reply #4
IMO a 5-10 kbps difference at -V2 isn't much of a concern, especially as it's about the difference between 3.99 and 3.97b here. There have always been minor differences in average bitrate when Lame versions changed.
As far as I can see with a quality level around -V2 average bitrate didn't change much when going from 3.98 to 3.99. At -V5 or similar as well as -V0 or similar it's another story.
lame3995o -Q1.7 --lowpass 17

Lame 3.99.5 outputs smaller files than 3.97 ?

Reply #5
If memory serves me, that difference will likely be largely due to the improvement in the handling of 'silence'.

Lame 3.99.5 outputs smaller files than 3.97 ?

Reply #6
Its the 'fix' for sfb21 bloating. Forcing -Y on both 3.98 and 3.99 will show very similar bitrate.

Lame 3.99.5 outputs smaller files than 3.97 ?

Reply #7
I recently re-ripped a CD (can't remember which one now. I think it was an Ozzy album) to test a new EAC install, and I setup 3.99.5 to encode it with V2. I had originally encoded it years ago with 3.90.3 aps. It's amazing how much smaller the files are. Some are 30kbps smaller. At first I thought I did something wrong, but after some reading, it looks like everything is ok. I didn't do any ABXing, but I ripped a few more albums and listened and didn't notice any artifacts. LAME has come a long way.

Lame 3.99.5 outputs smaller files than 3.97 ?

Reply #8
If one encodes aps -Y and V2 -Y you see a similar bitrate. It is a new handling of the sfb21 and it looks like a big improvement for lame. I checked the a standard V2 encode and the spectrum analysis (above 16khz) looked fine just like old versions.


Lame 3.99.5 outputs smaller files than 3.97 ?

Reply #10
What sample is the blue circle which is encoded significantly worse with 3.99.5 compared to 3.98 and 3.97?
lame3995o -Q1.7 --lowpass 17


Lame 3.99.5 outputs smaller files than 3.97 ?

Reply #12
It is a new handling of the sfb21 and it looks like a big improvement for lame.

Not sure about big improvements.


I think shadowking was referring not to sound quality changes but to bitrate improvements at roughly constant quality (and it's constant quality that VBR such as -V2 is trying to achieve), that have been achieved with some clever sfb21 handling (but largely disappear when using -Y, as that switch over-rides sfb21 bloat by ignoring accuracy requirements of the psymodel in that 16kHz+ band - though the frequencies are still present)

The error bars appear to support a Null Hypothesis that "there is no difference in average quality between -V2.5new and -V2.3old" over Kamedo2's test corpus to Kamedo2's ears, though specific samples vary noticeably. This doesn't speak to comparisons versus 3.90.3 --alt-preset standard, which is what shadowking referred to.

Thanks for pointing that out, by the way, shadowking. I hadn't realised that such improvements had been made in sfb21 handling over the years and thus in bitrate when -Y is not used.
Dynamic – the artist formerly known as DickD