Re: software speech problem after latest also upgrade

[Gregory Nowak <greg@romuald.net.eu.org>]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Hi Chuck and all.

On Tue, Mar 04, 2008 at 08:06:36PM -0500, Chuck Hallenbeck wrote:
> On my system, a similar problem with software speech followed the
> upgrade of the alsa libraries (libasound2) from 1.0.15 to 1.0.16, and
> takes the form of the disappearance of the first couple hundred
> milliseconds of a wave file played by aplay.

I don't know if libasound2 specifically was one of the packages
upgraded during that upgrade, but if there were some packages with
alsa in their name, it's likely that libasound2 was in that batch as
well. What I can find out, and have found out is that currently, the
libasound2 on my systems is at 1.0.16-1. Having done some testing with wav
files, I do see that a bit of the beginning does seem to be cut on
some, but not on all of them, and when I say all of them, I only mean
those where the sound starts literally playing as soon as the file starts
playing. This doesn't seem limited to aplay only as you mentioned
below, because it also does show up in wav files played by mplayer,
which is also using alsa of course.

> The problem does a real number of software speech with espeak, which is
> made up of numerous very short wave files, so it is more noticeable.

I'm confused here. My impression was that the espeak module (I mean
espeak, not espeak-generic) provided with speech-dispatcher did away
with generating wav files, and espeak send the text to be spoken to
the audio hardware directly, rather than generating a wav file, which
was later played back, as is the case in the espeak-generic module. Am
I wrong on this?

> When you use flite, each wave file produced by that engine contains
> leading and trailing silent periods, which makes it seem more sluggish,
> but protects you from the alsa problem.

That doesn't seem like it is the case. As I said in my previous post,
the problem is less noticeable in flite, but it's still there, it isn't
completely gone.

> I wish I knew how to backtrack to 1.0.15, but that version of the alsa
> libraries is no longer available in the Debian unstable archive.

If the wav file issue is in fact what is causing the problem, then I
would see backtracking to libasound2 1.0.15 as a temporary fix, and
not a desirable one in the long run. I think a better solution would
be to see if it is possible to make espeak generate a few-hundred ms
of silence in wav files it outputs, assuming that the espeak module
still relies on wav files, and that this is the problem.

Thanks for the info Chuck.

Greg


- -- 
web site: http://www.romuald.net.eu.org
gpg public key: http://www.romuald.net.eu.org/pubkey.asc
skype: gregn1
(authorization required, add me to your contacts list first)

- --
Free domains: http://www.eu.org/ or mail dns-manager@EU.org
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.6 (GNU/Linux)

iD8DBQFHzhVo7s9z/XlyUyARArsBAJ9h57oR4IXunZqf6S816JlRtIZcGACdFuFX
eXorYFbdMOv63JUO/nqxcbA=
=znG3
-----END PGP SIGNATURE-----