Comments on the Text to Speech "algorithm"

[William Hubbs]

On Sun, Nov 14, 2010 at 06:26:24PM +0000, Christopher Brannon wrote:
> Gilles Casse <gcasse at oralux.org> writes:
> 
> > - 78000 pre-recorded words,
> > - 5GB of disk space, wav files, 44100Hz,
> > - Each word is a separate wav file having the same name as the word in
> >   the text.
> 
> What happens if those WAV files are encoded as 32 kbit/s OGG/Vorbis or
> MP3 files?  The disk space requirement would shrink to approximately 120
> megabytes.  The most-frequently-used words could be cached in memory.
> Let's suppose a cache of 10000 words, for a memory footprint of 12
> megabytes.  That's acceptable on modern hardware, and it should be
> fairly responsive.
> 
> I still don't see the appeal of this technique, but to each his own.

Chris, I have to agree.  I definitely do not see the appeal of this
technique.

I'm no expert, but I think you will lose quality as you start to get
faster due to compressing the recording to make it respond faster.

Not only that, but according to the wikipedia article on the English
language [1], 78000 words does not even scratch the surface of the
number of words in English.

William

[1] http://en.wikipedia.org/wiki/English_language
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 198 bytes
Desc: not available
URL: 
<http://lists.freebsoft.org/pipermail/speechd/attachments/20101114/5be718b2/attachment.pgp>