[hts-users:01754] Was HVite modified much?
- Subject: [hts-users:01754] Was HVite modified much?
- From: Tienshyang Huei Liang <tshlmail-hts@xxxxxxxxx>
- Date: Thu, 30 Oct 2008 03:29:51 -0700 (PDT)
- Delivered-to: hts-users@xxxxxxxxxxxxxxx
- Domainkey-signature: a=rsa-sha1; q=dns; c=nofws; s=s1024; d=yahoo.com; h=X-YMail-OSG:Received:X-Mailer:Date:From:Reply-To:Subject:To:MIME-Version:Content-Type:Content-Transfer-Encoding:Message-ID; b=o7wapBjsx2SjGy/+EdA0K7wJxroP30IfVr6lYUkIOqW5aTPmU6D08wW0ZLk8qcJpWomBejA1nGiu8ahT07FgOR6Tq+Rg/bqh1HWQ5IFA5emJ+iZuDAMXY4wUaINsrsn168HzaOqOQ790aQmCSiMiIFBbE27vOMNB7Sy3PBJKIHs=;
Hi, HTS Working Group,
Sorry for this is actually a question regarding recognition rather than synthesis. I am trying recognizing speech by HVite with an SI model set and adaptation transform matrices, both of which were trained by HTS 2.1. However, no matter whether I gave the transform matrices to HVite, I got the same recognition results. That is, the transform matrices didn't take effect. What made me feel strange is that these transform matrices could take effect when synthesizing speech by HMGenS. So I wonder if you made any significant modification to either the original HVite or the format of transform matrices etc.
Of course it is very likely that I didn't use HVite correctly. I have converted HSMMs into HMMs before recognition. This is the command that I used:
/Tools_bin/HTS2.1/bin/HVite -A -C /HTS-demo_CMU-ARCTIC-ADAPT/configs/recog.cnf -D -T 1 -S /HTS-demo_CMU-ARCTIC-ADAPT/data/scp/recog.scp -p -75 -h */cmu_us_arctic_%%%_* -l /HTS-demo_CMU-ARCTIC-ADAPT/recog/qst003/ver1 -H /HTS-demo_CMU-ARCTIC-ADAPT/models/qst003/ver1/cmp/triphone_all.mmf -J /HTS-demo_CMU-ARCTIC-ADAPT/models/qst003/ver1/cmp/xforms SI+dec_feat3 -k -J /HTS-demo_CMU-ARCTIC-ADAPT/models/qst003/ver1/cmp/xforms/dec_tri.base -J /HTS-demo_CMU-ARCTIC-ADAPT/models/qst003/ver1/cmp/xforms/dec_tri.tree -w /HTS-demo_CMU-ARCTIC-ADAPT/data/dicts/mono.net /HTS-demo_CMU-ARCTIC-ADAPT/data/dicts/mono..dct /HTS-demo_CMU-ARCTIC-ADAPT/models/qst003/ver1/cmp/tiedlist_tri
Just by following the Section 3.6.3 (Step 14 - Evaluation of the Adapted System) of HTK Book 3.4, I turned on the switches "-k" and "-J" in order to involve matrices, trees and base classes for adaptation.
Thank you very much.
--
Hui LIANG