How to replace features in an .ark file?

[Anantha Krishnan]

I wish to use a different feature set, STE (Short Term Energy). After an .ark file with these features, I could use traditional Kaldi way of training GMM-HMM. (STE is used for EEG, not for speech here)

I wish to first decode the raw_mfcc_train.1.ark, replace the frame level features, and finally encode the same way. I understand that ark file is archive holding all key-value pairs. keys being utterance IDs and values being features per frame.

1. What encoding is used? 

2. How do I decode it using python's simple with open() function maybe in 'rb' mode?

[Guodong Ma]

If I understand correctly. Many tools can do these. For example: https://pypi.org/project/kaldiio/

Anantha Krishnan <ananthakrishn...@gmail.com> 于2024年3月14日周四 12:16写道：

--
Go to http://kaldi-asr.org/forums.html to find out how to join the kaldi-help group
---
You received this message because you are subscribed to the Google Groups "kaldi-help" group.
To unsubscribe from this group and stop receiving emails from it, send an email to kaldi-help+...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/kaldi-help/25bd7dd8-fa75-45b8-86ec-9c5dcaef983cn%40googlegroups.com.