EURASIP Journal on Audio, Speech, and Music Processing

Table 11 Language modeling complexity in terms of SPS. Lower SPS implies lower complexity

From: Improving speech recognition systems for the morphologically complex Malayalam language using subword tokens for language modeling

n-gram	Word	Morf.^a	BPE	Uni.^b	Syl.^c	S-BPE
	SPS
2	45	88	82	108	157	78
3	42	68	64	84	109	62
4	42	63	61	79	93	60
5	42	62	61	78	85	60
6	42	62	61	79	82	60

^a Morfessor
^b Unigram
^c Syllable

Back to article page