HMM automatic vowel detection

[ChenZheChina]

I also had to update the script to assign each letter to one of the states. I have lost this script in a computer crash about one year ago, so I don't remember the details, but of course this was based on the B matrix. The simplest thing would be to assign each letter to that of the two states where that letter has the highest weight.

Hi Marco,

Thanks for your reply.

I tried to print rdict (the mapping from indices to letters), newA (A Matrix) and newB (B Matrix) and got this for a small portion of English test data from Wikipedia:

Code:

======== rdict ======== {0: ' ', 1: 'a', 2: 'b', 3: 'c', 4: 'd', 5: 'e', 6: 'f', 7: 'g', 8: 'h', 9: 'i', 10: 'j', 11: 'k', 12: 'l', 13: 'm', 14: 'n', 15: 'o', 16: 'p', 17: 'q', 18: 'r', 19: 's', 20: 't', 21: 'u', 22: 'v', 23: 'w', 24: 'x', 25: 'y', 26: 'z'} ======== newA ======== [[0.209 0.791]  [0.647 0.353]] ======== newB ======== [[0.    0.    0.02  0.057 0.072 0.014 0.036 0.04  0.    0.    0.    0.008 0.109 0.043 0.172 0.    0.023 0.    0.103 0.13  0.125 0.    0.017 0.023 0.005 0.    0.001]  [0.287 0.152 0.001 0.    0.    0.163 0.003 0.036 0.071 0.116 0.    0.    0.    0.    0.    0.105 0.003 0.    0.    0.    0.004 0.043 0.    0.    0.    0.018 0.   ]]

rdict was not in the code. I added it. There was only cdict which maps from letters to indices:

Code:

cdict = dict([(x, i) for i, x in enumerate(chars)])

But it is easy to add a reversed one:

Code:

rdict = dict([(i, x) for i, x in enumerate(chars)])

Then we could assign letters to states where they have the maximum value:

Code:

highest_indices = np.argmax(newB, axis=0) sets = dict([(i, []) for i, _ in enumerate(newB)]) for i, s in enumerate(highest_indices):     sets[s].append(rdict[i])

In this case, if you want to assign letters based on minimum values, you could simply replace np.argmax with np.argmin.

Finally, we print sets, and get:

Code:

======== sets ======== {0: ['b', 'c', 'd', 'f', 'g', 'j', 'k', 'l', 'm', 'n', 'p', 'q', 'r', 's', 't', 'v', 'w', 'x', 'z'], 1: [' ', 'a', 'e', 'h', 'i', 'o', 'u', 'y']}

If I am not mistaking anything, this is how you get the “sets”, right?

(the attachment is some English test data, copied from Wikipedia)