Algorithm 2: Dynamic time warping.

1: Input:

2: $st$: template segment, with $n_{st}$ sampling points $s{t}_1$ to $s{t}_{n_{st}}$.

3: s: segment for recognition, whose class is unknown, with $n_s$ sampling points $s_1$ to $s_{n_s}$.

4: Output:

5: d: DTW distance between s and $st$.

6: normalize the amplitude of each sampling point in s into [-1, 1];

7: normalize the amplitude of each sampling point in $st$ into [-1, 1];

8: $d_{00}$ = 0;

9: for i = 1 to $n_{st}$ do

10:  $d_{i0}$ = +∞;

11: end for

12: for j = 1 to $n_s$ do

13:  $d_{0j}$ = +∞;

14: end for

15: for i = 1 to $n_{st}$ do

16:  for j = 1 to $n_s$ do

17:   ${\tilde{d}}_{ij}$ = $|s{t}_i-s_j|$;

18:   $d_{ij}$ = ${\tilde{d}}_{ij}$+ min$(d_{(i-1)j}$, $d_{(i-1)(j-1)}$, $d_{i(j-1)})$;

19:  end for

20: end for

21: d = $d_{n_{st}{n}_s}$;