I downloaded "4000 senone, 64 Gaussian continuous density models (for Sphinx-3)" and adapted it by following "Adapting the Acoustic Model (Sphinx3 version)". The adaptation data were recorded from telephone line sampled at 8000Hz. The corpus (about 500) contains limited (<10) simple phrases , such as 'mockbock', 'duck', 'voicenote'.
Then I decoded the testing samples with the adapted model, by adding -mllr mllr_matrix.
I downloaded "4000 senone, 64 Gaussian continuous density models (for Sphinx-3)" and adapted it by following "Adapting the Acoustic Model (Sphinx3 version)". The adaptation data were recorded from telephone line sampled at 8000Hz. The corpus (about 500) contains limited (<10) simple phrases , such as 'mockbock', 'duck', 'voicenote'.
Then I decoded the testing samples with the adapted model, by adding -mllr mllr_matrix.
Error msg:
ERROR Fr 0, best HMM score > 0 (2147436502); int32 wraparound?
ERROR Fr 2, best HMM score > 0 (2147482729); int32 wraparound?
ERROR Fr 3, best HMM score > 0 (2147339147); int32 wraparound?
ERROR Fr 4, best HMM score > 0 (2147338212); int32 wraparound?
....
...
Anybody give any suggestion or guide? your help is appreciated.
Most likely your test data has zero energy regions and you need to add "-dither yes" to solve this issue
May i know the procedure followed for speaker adaptation training..
http://www.speech.cs.cmu.edu/cmusphinx/moinmoin/AcousticModelAdaptation
I followed the procedures under the section "Adapting the Acoustic Model (Sphinx3 version)"