Study about Chinese Speech Synthesis Algorithm and Acoustic Model Based on Wireless Communication Network

Table 4

Changes in system recognition rate and model training rate after parallel optimization.

Experimental configuration	WER/%			Speed/(words·s^-1)	Speed increase multiple
	n-gram	RNN	RNN+n-gram
Baseline (CPU),	23.45	24.5	22.08	468	_
,	23.45	25.6	21.66	859	2.12
,	23.45	25.6	22.04	18466	37.56
,	23.45	26.5	23.14	23451	39.54