Newsgroups: comp.speech
Path: cantaloupe.srv.cs.cmu.edu!das-news2.harvard.edu!news2.near.net!howland.reston.ans.net!agate!news.ucdavis.edu!csus.edu!netcom.com!ebohlman
From: ebohlman@netcom.com (Eric Bohlman)
Subject: Re: Language dependent
Message-ID: <ebohlmanD2ruAK.KoF@netcom.com>
Organization: OMS Development
X-Newsreader: TIN [version 1.2 PL1]
References: <3frlvv$n8n@news.kth.se>
Date: Sat, 21 Jan 1995 20:01:32 GMT
Lines: 18

Patrik Johansson (pj@linnea-grind.stacken.kth.se) wrote:
: Am i right if i am assuming that a speech recognition system that has to be 
: trained is independent of the language spoken?

: I am talking about were you have to learn it every word, not just "get
: to know better" systems.

Not necessarily.  It would depend on what speech features were used in 
the matching.  For example, if the feature extraction process stripped 
out information about pitch slopes, the system wouldn't work well for 
languages (such as Chinese dialects) where rising or falling tones are 
used to distinguish words.  Some African languages include click-like 
sounds which might be poorly represented in systems intended for use with 
Western languages.

Also, the methods used for disambiguation are often based on the 
grammatical structure and word frequencies of a particular language.

