hmm rows of transmat_ must sum to 1.0 (got [1. 0. 1. 1.])

[Nian-Chen]

First of all, thank you for open source

There is a bug in hmm.py. If you run hmm.py directly, you get 16 hmm_models for each number. The reason is that the code puts "model = hmm.GaussianHMM(n_components=4, covariance_type='diag', n_iter=1800)" into the second loop. It should be moved to the first layer of the loop

[Sherry-XLL]

@Nian-Chen

Thanks very much for your nice corrections. In the first version of the code, I only implemented DTW and forgot to update GMM and HMM after implementing them. I have fixed bugs and uploaded the latest version of the code, and thank you again for your attention.

[Nian-Chen]

Hi, thanks for your reply, but it seems unreasonable to assemble the MFCC features of multiple samples to train a model. In fact, you can pass in the parameter "lengths" in "model.Fit", which can continuously update the model with multiple samples

Take the digit_0 as an example:
model.fit(X, lengths=X_lenth)
X_lenth is like [20 13 27 17 8 23 17 26 27 16 27 7 25 20 23 25]

[Baron-smith]

Your zip file has been damaged , could you send me a complete file to me to 13561364701@163.com. Thanks