Laser & Optoelectronics Progress, Volume. 57, Issue 18, 181702(2020)
Speaker-Dependent Speech Recognition Algorithm for Laparoscopic Supporter Control
Figures & Tables(9)
Fig. 5. Diagrams of i-vector parameter fusion and adding rejection identification unit. (a) Parameter fusion of i-vector; (b) adding rejection identification unit
Table 1. LSTM recurrent neural network model structure with i-vector feature
View tableTable 1. LSTM recurrent neural network model structure with i-vector feature
Layer ID Name Number of units Activation function 1 Input layer - - 2 FC1 64 ReLU 3 FC2 64 ReLU 4 FC3 64 ReLU 5 BiLSTM 64 - 6 FC4 64 ReLU 7 FC5 64 ReLU 8 Output layer - Softmax
Table 2. Recognition results of surgeon speech by three models
View tableTable 2. Recognition results of surgeon speech by three models
Word ID DTW GMM-HMM LSTM RNN with i-vector Total Correct Error Total Correct Error Total Correct Error FR FA FR FA FR FA 1 60 50 4 6 60 53 1 6 60 59 1 0 2 60 54 2 4 60 54 2 4 60 60 0 0 3 60 54 2 4 60 56 1 3 60 60 0 0 4 60 53 2 5 60 55 1 4 60 59 1 0 5 60 52 3 5 60 54 1 5 60 60 0 0 6 60 54 3 3 60 55 3 2 60 60 0 0 7 60 52 6 2 60 52 1 7 60 60 0 0 8 60 54 2 4 60 54 1 5 60 60 0 0 Sum 480 423 24 33 480 433 11 36 480 478 2 0
Table 3. Recognition results of assistant doctors speech by three models
View tableTable 3. Recognition results of assistant doctors speech by three models
Word ID DTW GMM-HMM LSTM RNN with i-vector Total Rejection FA Total Rejection FA Total Rejection FA 1 60 53 7 60 54 6 60 60 0 2 60 55 5 60 56 4 60 60 0 3 60 58 2 60 57 3 60 60 0 4 60 54 6 60 55 5 60 60 0 5 60 56 4 60 56 4 60 60 0 6 60 53 7 60 52 8 60 60 0 7 60 55 5 60 56 4 60 60 0 8 60 56 4 60 57 3 60 60 0 Sum 480 440 40 480 443 37 480 480 0
Table 4. Recognition results of interference speech by three models
View tableTable 4. Recognition results of interference speech by three models
Word ID DTW GMM-HMM LSTM RNN with i-vector Toatl Rejection FA Total Rejection FA Total Rejection FA 1 80 72 8 80 70 10 80 80 0 2 80 72 8 80 75 5 80 77 3 3 80 75 5 80 76 4 80 78 2 4 80 73 7 80 72 8 80 76 4 5 80 73 7 80 74 6 80 77 3 6 80 74 6 80 75 5 80 78 2 7 80 75 5 80 75 5 80 79 1 8 80 73 7 80 72 8 80 79 1 Sum 640 587 53 640 589 51 640 624 16
Tools
Get Citation
Copy Citation Text
Kailong Ren, Yi Wang, Xiaodong Chen, Huaiyu Cai. Speaker-Dependent Speech Recognition Algorithm for Laparoscopic Supporter Control[J]. Laser & Optoelectronics Progress, 2020, 57(18): 181702
Paper Information
Category: Medical Optics and Biotechnology
Received: Feb. 5, 2020
Accepted: Mar. 19, 2020
Published Online: Sep. 2, 2020
The Author Email: Wang Yi (koala_wy@tju.edu.cn)
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20