help
Closed this issue · 18 comments
I don't quite understand the methods you have given. Can you give me a basic demo, thank you!
Could you be more specific as to which method you are referring to?
您可以更具体地说明您指的是哪种方法？
I looked at the example given by kaldi/bable, but I don't quite understand how to connect ASR with kws. I want to create a small example, such as doing an example of a keyword on TIMTI.
您可以更具体地说明您指的是哪种方法？
Since I found the most basic demo on the Internet, I really don't quite understand how to get started. I am writing a master's thesis, but for the experiment I don't know how to use the KWS experiment. Can you guide me? Thank you!
Assuming you have a kaldi-based ASR setup, you need three files for setting up your KWS system
- ecf file
- keyword list
- rttm file
Have you tried creating these already? This is also described in: http://kaldi-asr.org/doc/kws.html
You can create each of these files using the commands given in this repo.
假设您有基于kaldi的ASR设置，则需要三个文件来设置KWS系统
- ecf文件
- 关键字列表
- rttm文件
你有没有试过创建这些？这也在以下描述：http：//kaldi-asr.org/doc/kws.html
您可以使用此repo中给出的命令创建每个文件。
假设您有基于kaldi的ASR设置，则需要三个文件来设置KWS系统
- ecf文件
- 关键字列表
- rttm文件
你有没有试过创建这些？这也在以下描述：http：//kaldi-asr.org/doc/kws.html
您可以使用此repo中给出的命令创建每个文件。
假设您有基于kaldi的ASR设置，则需要三个文件来设置KWS系统
- ecf文件
- 关键字列表
- rttm文件
你有没有试过创建这些？这也在以下描述：http：//kaldi-asr.org/doc/kws.html
您可以使用此repo中给出的命令创建每个文件。
My kaldi ASR model is kaldi/egs/thchs30.
I created exf.xml according to the command given by your repro, and I created kwlist.xml manually. However, for rttm I don't know how to create it.
Could you be more specific as to which method you are referring to?
I know that you are very busy, can you tell me your email address? This way I have a problem and can email you very quickly.
To create the rttm, you can use the script create_rttm.sh. It has two steps, first to create "ali" directory and then create rttm file from it.
to create "ali" dir: you will need set two inputs in the create_rttm.sh file:
$feats and $lang. $feats is the location of the audio features and related files generated from the dataset search on. $lang is the directory where your L.fst is generated.
I would prefer to answer your queries, as it might help somebody else as well. If you want you can obtain my email from my webpage mittulsingh.com
To create the rttm, you can use the script create_rttm.sh. It has two steps, first to create "ali" directory and then create rttm file from it.
to create "ali" dir: you will need set two inputs in the create_rttm.sh file:
$feats and $lang. $feats is the location of the audio features and related files generated from the dataset search on. $lang is the directory where your L.fst is generated.I would prefer to answer your queries, as it might help somebody else as well. If you want you can obtain my email from my webpage mittulsingh.com
What is the role of .common/slurm_dep_graph.sh in create_rttm.sh, I didn't find this file
You can find this file below, this is used to run jobs onto a cluster managed using slurm utility. https://github.com/psmit/kaldi-recipes/blob/master/common/slurm_dep_graph.sh
You can choose to remove this script as well and edit the create_rttm.sh script accordingly.
The rrtm file generated by running ali_to_rttm.sh in your report is blank.
I don't know how to start editing and re-editing create_rttm.sh
If you are not busy right now, can you remotely operate my computer through the teamViewer software?
Sorry, my connection is slow to be able to do that. What does your log/L_align.log say? This log folder will exist in the folder you got your rttm.
FATAL: FstCompiler: Symbol "data/lang/local//lexicon.txt" is not mapped to any integer arc ilabel, symbol table = data/lang/phones.txt, source = standard input, line = 4
ERROR: FstHeader::Read: Bad FST header: standard input
My command input is sudo kws_scripts/ali_to_rttm.sh data/train data/lang exp/tri4b_dnn_ali
Seems like the file data/lang/phones.txt is in an incorrect format, specifically line 4. Hopefully, using the correct version of this file will correct the error. I must say I can only with the problems related to kws_scripts. ali_to_rttm.sh happens to be a Kaldi repo file and is included in this repo just as an example. I would suggest posting to Kaldi help forum if the issue persists.
Maybe my data set is Chinese, so I can't match it.
I would think not think that to be the case. For now, closing this issue.
@ckn123 i have meet the same question as you ,and i have solved it .
the key is in file ali_to_rttm.sh make_L_align.sh have 3 paras,the fisrt param should be data/local/lang_tmp !!!!