[hts-users:02274] Re: Why do we use log f0 instead of f0?

Subject: [hts-users:02274] Re: Why do we use log f0 instead of f0?

From: Zhijie Yan <zhijiey@xxxxxxxxxxxxx>

Date: Fri, 6 Nov 2009 08:49:27 +0000

Accept-language: en-US

Delivered-to: hts-users@xxxxxxxxxxxxxxx

Thread-index: AQHKXqGi0S7ZSbJKB0qMWnHNDwDVGJEovRzw

Thread-topic: [hts-users:02272] Why do we use log f0 instead of f0?

One reason is because we use Gaussian distribution to model the acoustic features in HTS. It was found in some literature that the distribution of log f0 could fit the model better. This is also true for some features, such as energy / gain.

Zhijie

From: jangwon kim [mailto:jangwonjkim@xxxxxxxxx]
Sent: Friday, November 06, 2009 1:16 PM
To: hts-users@xxxxxxxxxxxxxxx
Subject: [hts-users:02272] Why do we use log f0 instead of f0?

HTS basically use log f0, not f0, for excitation parameter.
I am curious about the reason for that.

Is there anyone can tell me why?

Thank you in advance.

Jangwon