home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!mcsun!uknet!bcc.ac.uk!link-1.ts.bcc.ac.uk!ubacr45
- From: ubacr45@ucl.ac.uk (Mr G Toal)
- Newsgroups: comp.speech
- Subject: Re: better phoneme directory for scat on SPARC
- Keywords: scat phoneme speech
- Message-ID: <1992Dec13.044503.26573@bas-a.bcc.ac.uk>
- Date: 13 Dec 92 04:45:03 GMT
- References: <7619@stan.xx.swin.oz.au>
- Sender: news@ucl.ac.uk (Usenet News System)
- Organization: Bloomsbury Computing Consortium, London
- Lines: 20
-
- In article <7619@stan.xx.swin.oz.au> quentin@swin.oz.au writes:
- >I am running scat on my SUN SPARC. Scat will speak stdin for you.
- >It uses a directory of phonemes, which are short (.1 to .3 s) .au files.
- >The trouble is that the speech sounds garbled. I believe that the problem
- >lies in the recording of the .au files for each phoneme. I tried to record
- >my own with some success but I was wondering if anyone has a directory
- >of phonemes for scat which makes the speech understandable.
-
- I don't think you'll ever get anything very impressive: it's a general
- limitation of the simple sampled-phoneme algorithm: phonemes simply
- don't join up. A marginally cleverer idea is to sample pairs of
- phonemes, and provide samples that run from the middle of one
- segment to the middle of the next. This means of course that you
- need N^2 samples instead of N (slightly less if you skip ones that
- don't happen in your language) This gives better speech because
- most of the rising or lowering inflexions are at the start and end
- of phonemes, and the middles are relatively stable and can be abutted
- with less noticable discontinuity.
-
- G
-