EURASIP Journal on Audio, Speech, and Music Processing

Table 10 Recognition results on the evaluation set using simple MSD models for POS tags (P) and MSD tags (M)

From: Context-dependent factored language models

First pass	Second pass	WER [%]	Relative
model	model		improvement[%]
Bigram	M	25.84	+11.56
Bigram	M+P	26.01	+10.99
Trigram	M	25.86	−0.84
Trigram	M+P	25.62	+0.07

Back to article page