Here are some interesting comments from Timothy Spencer I've been meaning to repost to mp3encoder: Date: Mon, 13 Dec 1999 23:42:17 -0800 From: "T. Spencer" <[EMAIL PROTECTED]> X-Accept-Language: en To: [EMAIL PROTECTED] Subject: Re: LAME Content-Type: text/plain; charset=us-ascii X-UIDL: PCE!!@9Qd9$V(e9A5;!! X-UID: 794 Mark, Sure, I'd like to help. You can currently read the bulk of my critique at mp3.com's bulletin board under "general mp3 questions". This critique also stands up for LAME's 3.57. I used the Lame Batch frontend with 3.57 to encode wav files which were ripped from CD using Music Match set on multipass with digital correction which gives a high quality method for maintaining strong data integrity. I listen to the encoded mp3's and wav files back to back. I use Sonique which recognizes my Prodif 32 digital I/O card. The digital output is studio grade and the Stardust mp3 decoder is unarguably the best available. I also use my 2-way 50 watt PC speakers to listen as a 2nd opinion. Two different speakers give two different ways to monitor the quality. The first and most important difference between LAME and the wav files are the high frequency differences in volume. LAME sounds like it's increasing high frequency volume in certain frequency areas. I took an mp3 converted to wav and used soundforge to deemphasize the highs to good effect. A very good test track comes off of Tears for Fears - Elemental CD. Track 2 in particular titled "Cold". It has a high overall volume with many different sounds in the mix both in ambience and sounds that are in the front of the stereo image. It also has ambient high frequencies which are rare in many songs. I tried "Cold" on LAME 3.51 and needed to use the VBR-2 setting in stereo mode in order to come close to the CD. The average bitrate was around 232 kb/sec in order to achieve virtual transparency. The highs were still audibly emphasized compared to the wav, but the high frequency ambience finally sounded very close. Also, the lower midrange in the vocals were more filled out. Another difficult track to encode is found on Tori Amos' new CD titled "To Venus and Back" Disc 1. Track 1 called "Bliss". Listen for her inhale. Every time she inhales, there is a breakup on that sound. The original wav reveals that her voice has a certain effect on it, so when she inhales, it is not quite natural, but the LAME codec doesn't encode it correctly. No matter what bitrate I used, the result was the same on the "inhale". This problem also occurs on her CD single of Hey Jupiter where she sings live. Again, her inhales are slightly artifacted, even though on the wav, it's merely a microphone noise. LAME intensifies the microphone distortion and makes it stick out unnaturally. It sounds broken up slightly even at the highest bitrates. Here are two things I think LAME needs to improve on: 1. Highs need to be deemphasized especially at lower bitrates due to how shrill it sounds. 2. Lower midrange in the 250-800 Hz range needs to sound fuller. I get the impression much of this sound is missing when comparing the mp3 to the wav. I'm willing to bet LAME and most other encoders sacrifice this area and concentrate more on the 2,000-4,000 Hz where human hearing is very sensitive and hears distortion more easily. Even Fraunhofer sounds thin in the lower midrange but I get the impression they hide this by making their codec sound "fluffy" on much of the sound regardless of bitrate. Fraunhofer's highs also seem more restrained, less dynamic. I feel LAME has something that Fraunhofer does not currently have. Better dynamics. So, at this point, I don't think you should try to be more like Fraunhofer, but instead try to go your own route, with your own signature. I'm willing to bet you'll surpass Fraunhofer with some more work on your perceptual model. You may even be able to develop a wav modifier which can apply specially chosen equalization and adjustment before encoding is done. This way you can better shape the quality of your resulting mp3. I've found that modifying wavs with mastering programs before encoding can help bring a result closer to the original wav than was possible before. This could be a new twist on your encoder. With the right programming, you could have your encoder modify the digital wav or CD data just before it's encoded, with only minimal time added to the encoding process. You could literally have choices for the different DSP effects. This could help someone get a better result even at lower bitrates. The sound might even be different from the wav, but better sounding given the bitrate used. I've found also that when I change eq or stereo ambience, the bitrates in VBR also change. Sometimes higher, and sometimes lower. This is why I believe it's possible to alter the sound before encoding to increase transparency in the result. I hope this is the type of input you are looking for. Please let me know if I'm helping with this kind of information. Timothy -- MP3 ENCODER mailing list ( http://geek.rcc.se/mp3encoder/ )