Newsgroups: comp.speech
Path: cantaloupe.srv.cs.cmu.edu!das-news2.harvard.edu!news2.near.net!howland.reston.ans.net!gatech!udel!news.mathworks.com!news.kei.com!travelers.mail.cornell.edu!newstand.syr.edu!galileo.cc.rochester.edu!uhura.cc.rochester.edu!rsra
From: rsra@uhura.cc.rochester.edu (Mike McCourt)
Subject: Re: Evaluating Synthesised Speech
Message-ID: <1994Nov11.130450.28362@galileo.cc.rochester.edu>
Sender: news@galileo.cc.rochester.edu
Nntp-Posting-Host: uhura.cc.rochester.edu
Organization: University of Rochester - Rochester, New York
References: <39m10b$khj@mercury.dur.ac.uk> <784498684snz@lfheller.demon.co.uk>
Date: Fri, 11 Nov 94 13:04:50 GMT
Lines: 32

In <784498684snz@lfheller.demon.co.uk> Leon@lfheller.demon.co.uk (Leon Heller) writes:

>In article <39m10b$khj@mercury.dur.ac.uk>
>           I.H.Morgan@durham.ac.uk "I.H Morgan" writes:

>> Hi !
>>  I'm doing some work on copy synthesis, and I will need to compare some
>> synthesised speech with the original speech sample. 
>>  I would therefore be interested in any information on, or references to,
>> methods for using human listeners to quantitatively compare a sample of
>> synthesised speech, with either the original human speech, or another
>> synthesised sample, (of the same words.) (I'm thinking of considering overall
>> `intelligibility', number of words correctly recognised, and simple preference
>> between two samples, but I'd be interested in other ideas, and in the
>> statistics needed to analyse the results.)
>>  Many thanks,
>> Ivor Morgan

>DEF STAN 00-25 Part 9 (published by the MOD) has something on this. It's
>not restricted, and your library should be able to get hold of it. The
>Modified Rhyme Test and Diagnostic Rhyme Test are recommended. The DRT
>is covered in Speech Intelligibility and Speaker Recognition, (Ed) Hawley,
>ME, Dowden, Hutchinson and Ross, Inc. Stroudsberg, USA.

>Leon

Is there any approach to do this process by machine and not by humans, e.g.
comparing what somebody says to a model and then making the machine decide
if the speaker said it correctly or not (pronunciation in a foreign language
for example)?

Alex
