Really should do this. It will almost surely help. It will be a decently complex implementational effort, but at least the HTK book is a reliable guide.
If we're going down this route, it may be worthwhile to appraise whether it would be better to substitute the Kaldi toolkit for HTK, since it is mostly compatible, much faster, and actively developed. (It also has a much more permissive license, so we wouldn't have to send people complex directions for installing HTK; we could just build whatever parts of Kaldi we need right into Prosodylab-Aligner) This would make it easier to import advancements in ASR (since 1995 or so) into Prosodylab-Aligner; for instance, Kaldi even supports deep neural network acoustic models (which are obsoleting triphones as we speak).
Really should do this. It will almost surely help. It will be a decently complex implementational effort, but at least the HTK book is a reliable guide.
If we're going down this route, it may be worthwhile to appraise whether it would be better to substitute the Kaldi toolkit for HTK, since it is mostly compatible, much faster, and actively developed. (It also has a much more permissive license, so we wouldn't have to send people complex directions for installing HTK; we could just build whatever parts of Kaldi we need right into Prosodylab-Aligner) This would make it easier to import advancements in ASR (since 1995 or so) into Prosodylab-Aligner; for instance, Kaldi even supports deep neural network acoustic models (which are obsoleting triphones as we speak).