Re: flac 1.40 verses ffmpeg
"Michael D. Lawler" <[email protected]> Tue, 27 Sep 2022 11:10:04 -0400
| Newsgroups | gmane.comp.audio.compression.flac.user |
|---|---|
| Message-ID | <[email protected]> |
--===============2421081328865440829== Content-Type: multipart/alternative; boundary="=====================_449054890==.ALT" --=====================_449054890==.ALT Content-Type: text/plain; charset="iso-8859-1"; format=flowed Content-Transfer-Encoding: quoted-printable These are all 44.1kHz and adding -b 4608 helped,=20 but ffmpeg still does better than flac for this handful of files. At 09:07 AM 9/27/2022, Martijn van Beurden wrote: >Op di 27 sep. 2022 om 14:22 schreef Michael D.=20 >Lawler <<mailto:[email protected]>[email protected]>: >My conclusion is that ffmpeg's library is doing something that -e >does, but not using a lot of CPU so perhaps this could be studied and >flac could be updated to improve compression further with reasonable >CPU usage.=C2 I have all of the files and could provide the ones where >-e is necessary and the few where ffmpeg still does better with -8pe >--lax -l 32. > > > >One important difference between libflac and=20 >ffmpeg is that they use different blocksizes.=20 >FLAC as you've used it will default to blocksize=20 >4096 no matter what sample rate the input is.=20 >ffmpeg will adapt its blocksize to the=20 >samplerate. For 44.1kHz and 48kHz, it uses=20 >blocksize 4608. In the past it has been shown=20 >this benefits certain kinds of music, but the=20 >4096 used by libflac does better for others.=20 >ffmpeg uses a larger blocksize for higher=20 >samplerates and a smaller blocksize for lower=20 >samplerates. This might be the reason you're seeing this difference. > >Perhaps you can find out what the sample rates=20 >are of the files that ffmpeg compresses better=20 >than libflac. If those are mainly 44.1kHz or=20 >48kHz, you can try to improve on it with libFALC by adding -b 4608 > >Kind regards, Martijn van Beurden --=20 Michael D. Lawler email mailto:[email protected] --=====================_449054890==.ALT Content-Type: text/html; charset="iso-8859-1" Content-Transfer-Encoding: quoted-printable <html> <body> These are all 44.1kHz and adding -b 4608 helped, but ffmpeg still does better than flac for this handful of files.<br><br> At 09:07 AM 9/27/2022, Martijn van Beurden wrote:<br> <blockquote type=3Dcite class=3Dcite cite=3D"">Op di 27 sep. 2022 om 14:22 schreef Michael D. Lawler <<a href=3D"mailto:[email protected]">[email protected]</a>>:<br> <dl> <dd>My conclusion is that ffmpeg's library is doing something that -e <br> <dd>does, but not using a lot of CPU so perhaps this could be studied and <br> <dd>flac could be updated to improve compression further with reasonable <br> <dd>CPU usage.=C2 I have all of the files and could provide the ones where <br> <dd>-e is necessary and the few where ffmpeg still does better with -8pe <br> <dd>--lax -l 32.<br><br> </dl><br><br> One important difference between libflac and ffmpeg is that they use different blocksizes. FLAC as you've used it will default to blocksize 4096 no matter what sample rate the input is. ffmpeg will adapt its blocksize to the samplerate. For 44.1kHz and 48kHz, it uses blocksize 4608. In the past it has been shown this benefits certain kinds of music, but the 4096 used by libflac does better for others. ffmpeg uses a larger blocksize for higher samplerates and a smaller blocksize for lower samplerates. This might be the reason you're seeing this difference.<br><br> Perhaps you can find out what the sample rates are of the files that ffmpeg compresses better than libflac. If those are mainly 44.1kHz or 48kHz, you can try to improve on it with libFALC by adding -b 4608<br><br> Kind regards, Martijn van Beurden </blockquote> <x-sigsep><p></x-sigsep> -- <br> Michael D. Lawler<br> email <a href=3D"mailto:[email protected]" eudora=3D"autourl"> mailto:[email protected]<br> </a></body> </html> --=====================_449054890==.ALT-- --===============2421081328865440829== Content-Type: text/plain; charset="us-ascii" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit Content-Disposition: inline _______________________________________________ Flac mailing list [email protected] http://lists.xiph.org/mailman/listinfo/flac --===============2421081328865440829==--