Acoustic Model with Multiple Lexicon Types for Indonesian Speech Recognition

Table 14

The performance of the models using 46,550 validated utterances, for the four-combination lexicon type.

Lexicon type	GMM-HMM model (%WER)
	Monophone	Triphone DELTA + DELTA-DELTA	Triphone LDA + MLLT	Triphone (SAT)
1c-vocab-full	81.25	61.96	50.66	51.24
enmap-1c-vocab-full	80.99	60.88	50.08	50.83
enmap-kv-vocab-full	79.46	65.92	55.23	56.05
kv-vocab-full	76.75	65.27	55.52	56.05
4-combination	72.17	51.27	41.57	42.80