On Mon, 20 Sep 1999, Mathew Hendry wrote:
> Hi all,
>
> I've found a small sound segment that produces very unpleasant artifacts
> with LAME's VBR mode. The VBR improvements added in 3.26 (#define RH)
> seem to make it worse. The sample is very rich in loud "s" sounds, which
> apparenty cause ringing artifacts. These disappear only if a minimum
> bitrate of 160kbps or higher is imposed, in both joint/ms and full
> stereo. The sample is only about 3 sec long (687,892 bytes); is anyone
> interested in me uploading it somewhere, and if so, where?
I've found that *ANY* pure speech sample will do this to some extent or
another (I went out an bought a few BOOKS-ON-CD for VBR test cases).
> Also, I'd like to add support to LAME to allow the user to specify a
> mean (average) bitrate, and am wondering about the best way to go about
> this. The simplest way would appear to be to output the bitrate/quality
> data to a file in a first pass (without performing a full encode), and
> then run a second pass to generate a "bitrate profile", which could then
> be read in for the third, encoding pass. Does this sound reasonable?
> Perhaps a general way to specify bitrate on a frame-by-frame basis would
> be useful anyway (e.g. for manually assisted VBR encoding)...
My current work has been on setting up a VBR mode curve where it will
increase the quality for the slightest improvement at low bit rates, and
require bigger improvements at higher bitrates. You set quality by
adjusting a gamma value.
My next project will be 'windowed target bitrates' Where you can tell the
encoder:
I want no more then 224Kbps avg in a 1 sec window, no more then 160Kbps
in a 5 sec window. No bitrate less then 128K unless there is silence
(tuniable by a ratio from ATH to absolute silence).
How do people on the list feel about having a Lame config file which
specifies these rules as named 'profiles'.
For that I'm waiting for VBR to settle some.
--
MP3 ENCODER mailing list ( http://geek.rcc.se/mp3encoder/ )
