Date: Tue, 13 Jun 2006 11:11:10 -0400 From: Chuck Swiger <cswiger@mac.com> To: Howard Jones <howie@thingy.com> Cc: freebsd-questions@freebsd.org Subject: Re: [freebsd-questions] Scanning MP3 files for skips Message-ID: <448ED58E.7010402@mac.com> In-Reply-To: <448ECAB3.5090900@thingy.com> References: <7.0.1.0.2.20060613145307.023662b0@broadpark.no> <448EC879.1010603@mac.com> <448ECAB3.5090900@thingy.com>
next in thread | previous in thread | raw e-mail | index | archive | help
Howard Jones wrote:
> Chuck Swiger wrote:
[ ...ID'ing skips in music... ]
>> Sort of...GraceNote and a few other companies ("Shazam", seems to be
>> from India?) sell a service where music files can be fingerprinted and
>> identified. Good audio files ought to ID as what they are; bad music
>> files with skips or garbage will fail to ID.
>>
> Shazam (at least) works on a fragment the song. In the UK they provide a
> phone-based service, which only needs 20-30 seconds of clear music to
> identify a song.
Yeah, that's right. But there's also a fingerprinting tool which creates
something called QCF files (Qualcom something-or-other) which analyses the
entire song and should notice major skips or distortions better.
[ ... ]
> I'm also looking for a blip-detecting MP3 tool. I haven't had time to
> look at it yet, but I was going to try something like libmad on the
> assumption that somewhere internally it knows when it's only had half an
> frame of data, even if there is no CRC. That way, it'll work on any
> obscure music I have, without relying on some external giant database of
> correctness.
That's the rub of the matter: most home-grown tools are going to find it hard
to recognize a skip from Trent Reznor or a lot of rap music, or a dropout with
silences common in classic music, etc. :-)
With the external giant database of correctness, you've got something which
actually can tell that a song isn't correct, rather than making guesses, even
if you have to break your samples down into 10 to 30 second pieces and check
them all individually.
[ Again, I would try using the QCF fingerprinting tool instead of exhaustive
submatch checking. On the other hand, if you can manage to put together
something which can guess well, I'd be interested in seeing it... ]
--
-Chuck
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?448ED58E.7010402>