User Ratings
Rate This Project
Login To Rate This Project
User Reviews
-
I started bavieca for about a week and tried to build Indonesian ASR for news reading domain. The results is very good compare to julius and HTK (5K words LM; WER around 2-3%; untrained speaker). Bavieca is a very promising tool.
-
I started using BAVIECA one month ago with the help of various scripts kindly made available by Daniel Bolanos and I was able to reproduce quite good results on WSJ, almost reproducing Dani's original results. I am now porting BAVIECA to Italian and I just trained an Adult Speech ASR (read speech, APASCI-FBK corpus) and a Child Speech ASR (ChildIt-FBK) both using also some MLLR adaptation iterations. Results are FANTASTIC and essentially better than those previously obtained with similar toolkit (Cslu Toolkit, Sonic, Sphinx and Julius). As a simple example, PHONETIC RECOGNITION ERROR was only 15% with children speech. More results will be reported as soon as I'll develop new things (VTLN, MMI, ecc.) Indeed, BAVIECA was quite useful. MANY THANKS DANIEL Piero Cosi Director/Coordinator ISTC CNR - UOS Padova Istituto di Scienze e Tecnologie della Cognizione Consiglio Nazionale delle Ricerche Unita' Organizzativa di Supporto di Padova Via Martiri della liberta', 2 35137 Padova (ITALY) tel: +39 049 8271822 fax: +39 049 8271825 e-mail: piero.cosi@pd.istc.cnr.it skype: pierocosi
- Previous
- You're on page 1
- Next