Classification of heterogeneous text data for robust domain-specific language modeling

EURASIP Journal on Audio, Speech, and Music Processing

Table 7 Language model perplexity and performance of Slovak LVCSR system with different acoustic models

			APD1+APD2 250 h		APD1+APD2 250 h		APD1+APD2		APD1+APD2
	Text		(table mic.)		(close-talk mic.)		+PAR 340 h		+PAR+BN 520 h
PPL	classification		sp. adapt.: no		sp. adapt.: no		sp. adapt.: no		sp. adapt.: no
			eval. set: gender-bal.		eval. set: gender-bal.		eval. set: gender-bal.		eval. set: gender-bal.
	Weighting	Similarity	Acc %	Corr %	Acc %	Corr %	Acc %	Corr %	Acc %	Corr %
40.4302	Reference language model		91.84	93.08	93.61	94.51	94.36	95.13	94.06	94.89
36.0428	tf-idf	Bhattacharyya	92.44	93.64	93.99	94.85	94.70	95.46	94.36	95.13
35.9444		Jaccard index	92.46	93.65	93.97	94.85	94.72	95.47	94.37	95.16
38.1756		Jensen-Shannon	92.23	93.39	93.78	94.70	94.50	95.25	94.21	94.99
38.1289	Okapi	Bhattacharyya	92.17	93.34	93.77	94.65	94.61	95.34	94.27	95.02
39.9782		Jaccard index	92.10	93.31	93.60	94.54	94.48	95.21	94.11	94.89
39.2267		Jensen-Shannon	92.27	93.42	93.77	94.67	94.61	95.36	94.18	94.95
40.1325	Ltu	Bhattacharyya	91.86	93.12	93.57	94.51	94.42	95.16	94.05	94.87
40.1439		Jaccard index	91.87	93.12	93.56	94.50	94.40	95.16	94.04	94.87
40.1319		Jensen-Shannon	91.87	93.12	93.57	94.51	94.42	95.16	94.05	94.87