[hts-users:03955] Re: Intelligibility

[Simon King <Simon.King@xxxxxxxx>]

On 26 Nov 2013, at 08:21, Muhammad Shoaib Ali wrote:

> i have generated the speech using HTS. Now i want to calculate the intelligibility of my generated voice i.e the subjective intelligibility . how can i calculate "Speech Intelligibility Index (SII)" of my generated speech.
> 

For 'subjective intelligibility' you need to run a listening test; the protocol used in the Blizzard Challenge is a typical way to do this:

e.g., http://www.synsig.org/index.php/Blizzard_Challenge_2013

(this SynSIG site is down for maintenance, but should be back soon; in the meantime you can get the papers from festvox.org)

To calculate SII, which is not a subjective measure but a physical measurement directly on the signal, you need a software implementation of the ANSI standard, such as the ones from http://www.sii.to or elsewhere.

But you must exercise caution - SII is designed for (degraded) natural speech, not synthetic speech in clean conditions. It is not a replacement for a listening test. For some papers which examine its correlation with subjective intelligibility under various conditions, try

http://listening-talker.org/publications.html

but again note that the focus there was on speech degraded by the presence of additive noise, which may not match your situation.

regards,
Simon



-- 
The University of Edinburgh is a charitable body, registered in
Scotland, with registration number SC005336.