Author Affiliations
School of Information Engineering and Automation, Kunming University of Science and Technology, Kunming 650500, Yunnan , Chinashow less
Fig. 1. Steps of syllable segmentation algorithm
Fig. 2. Waveform and its upper envelope of speech signal
Fig. 3. Some envelopes formed by various interpolation methods
Fig. 4. Original speech waveform and envelope after interpolation. (a) Original speech waveform; (b) envelope after interpolation
Fig. 5. Filtered envelope and logarithmic envelope. (a) Filtered envelope; (b) logarithmic envelope
Fig. 6. Adjustment of extreme points on logarithmic envelope. (a) Extreme points on logarithmic envelope; (b) extreme points after single threshold processing; (c) extreme points after thresholding
Fig. 7. Chinese phonetic syllable segmentation under noisy environment. (a) Original speech; (b) distribution of extreme points on logarithmic envelope of speech without noise; (c) speech with noises (SNR is 5 dB); (d) distribution of extreme points on logarithmic envelope of speech with noise
Filter type | SNR /dB |
---|
20 | 15 | 10 | 5 |
---|
Chebyshev filtering | 79.6 | 78.3 | 77.1 | 69.4 | Bessel filtering | 81.4 | 80.6 | 79.3 | 74.8 | Butterworth filtering | 91.7 | 91.0 | 90.6 | 87.5 |
|
Table 1. Segmentation accuracy of different filters for different signal-to-noise ratio
Segmentation algorithm | Accuracy /% |
---|
Singularity index[15] | 26.9 | Local singularity[13] | 50.3 | Short-term energy[12] | 76.9 | Fractal dimension[14] | 82.3 | Method of this article | 92.1 |
|
Table 2. Segmentation accuracy of different algorithms without noise
Segmentation algorithm | SNR /dB |
---|
10 | 5 | 0 | -5 |
---|
Singular index[15] | 24.7 | 20.6 | 18.7 | 16.9 | Local singularity[13] | 49.3 | 47.7 | 42.6 | 39.1 | Short-term energy[12] | 69.6 | 60.3 | 58.4 | 54.2 | Fractal dimension[14] | 83.4 | 80.1 | 79.7 | 71.2 | Method of this article | 90.6 | 87.5 | 76.3 | 70.6 |
|
Table 3. Speech segmentation accuracy of different algorithms under different signal-to-noise ratio
Segmentation algorithm | Voice duration /s |
---|
4 | 6 | 8 | 10 |
---|
Singular index[15] | 1.3 | 2.2 | 3.9 | 5.1 | Local singularity[13] | 2.9 | 4.1 | 5.7 | 7.4 | Short-term energy[12] | 3.1 | 4.6 | 5.9 | 7.6 | Fractal dimension[14] | 1.9 | 2.7 | 4.5 | 6.0 | Method of this article | 2.5 | 3.7 | 5.4 | 6.9 |
|
Table 4. Segmentation time of different algorithms under different durations