online2-wav-nnet2-latgen-faster管道定制开发中参数优化

We are trying to build and use online2-wav-nnet2-latgen-faster inwindows, but since there is no pipeline “|” operations, which are used in the following command in linux: 我们试图在Windows中构建和使用online2-wav-nnet2-latgen-faster-faster，但是由于没有管道“ |”操作，在Linux中的以下命令中使用：

~/kaldi-online/src/online2bin/online2-wav-nnet2-latgen-faster --do-endpointing=false \
 --online=false \
--config=nnet_a_gpu_online/conf/online_nnet2_decoding.conf \
 --max-active=7000 --beam=15.0 --lattice-beam=6.0 \
  --acoustic-scale=0.1 --word-symbol-table=graph/words.txt \ 
nnet_a_gpu_online/smbr_epoch2.mdl graph/HCLG.fst "ark:echo utterance-id1 utterance-id1|" "scp:echo utterance-id1 ENG_M.wav|" \ ark:/dev/null

We try to search the code to find how we can format the command in windows, but have no idea. Can anyone encounter the same problem and give some suggestions? Thanks in advance. 我们尝试搜索代码以找到如何在Windows中格式化命令，但一无所知。任何人都可以遇到相同的问题并提出建议吗？提前致谢。

You'll probably have to use temporary files. 您可能必须使用临时文件。

I wander what kind of temporary files can be used or use wav-copy? I will check the code and find it out. 我徘徊可以使用哪种临时文件或使用wav-copy？我将检查代码并找出来。

I mean instead of using 我的意思是不使用

                  "ark:echo utterance-id1 utterance-id1|"

as an input, do something like 作为输入，做类似的事情

echo utterance-id1 utterance-id1 >foo

and use 和使用

ark:foo

as an input. 作为输入。

Thanks, dan. As suggest, we come out with the following cmd: 谢谢，丹。如建议的那样，我们得出以下cmd：

online2-wav-nnet2-latgen-faster --do-endpointing=false --online=true --config=conf/online_nnet2_decoding.conf --max-active=7000 --beam=11.0 --lattice-beam=6.0 --acoustic-scale=0.1 --word-symbol-table=model/words.txt model/final.mdl model/HCLG.fst "ark:data/spk2utt" "scp:data/wav.scp" "ark:null"
                    

The spk2utt file has the content of “utterance-id1 utterance-id1”, and the scp file has the content of of “utterance-id1 test.wav”. We find this cmd not work, may be due the problem of “output part”. It is found that there is no similar part like in the gmm case, such as ”ark,t:.\data\trans.txt ark,t:.\data\ali.txt”. How to format the output part? spk2utt文件的内容为“话语-id1话语-id1”，scp文件的内容为“话语-id1 test.wav”。我们发现此cmd不起作用，可能是由于“输出部分”的问题。发现没有类似gmm的部分，例如“ ark，t：。\\ data \\ trans.txt ark，t：。\\ data \\ ali.txt”。如何格式化输出部分？

I think now is the time to declare rule 1 of the support forum.Rule 1: If you say that there is an error without showing the specific error message, your question will not be answered. 我想现在是时候宣布支持论坛的规则1规则1：如果您说有错误而没有显示特定的错误消息，则不会回答您的问题。

OK, I see the attachment via the web interface. Usually a program just hanging means it is waiting for standard input; however, it is not clear to me why this program with those arguments would try to read from the standard input. You might have to run it in a debugger; there is a way in Visual Studio to run a particular
command line. I won't be able to help you further with that, though. 通常，仅挂起一个程序就意味着它正在等待标准输入。但是，我不清楚为什么带有这些参数的程序会尝试从标准输入中读取内容。您可能必须在调试器中运行它。在Visual Studio中，有一种方法可以运行特定的命令行。不过，我将无法进一步帮助您。

That program by default uses multiple threads at startup, while computing the normalizers in the iVector extractor. This was done to reduce startup time when there are multiple CPUs available. Funnily enough, on Linux this was previously a problem when using multi-threaded OpenBLAS, because by default it is not built to be thread-safe. This was fixed by the following line in tools/Makefile: 默认情况下，该程序在启动时使用多个线程，同时在iVector提取器中计算规范化器。这样做是为了减少有多个CPU时的启动时间。有趣的是，在Linux上，使用多线程OpenBLAS以前是一个问题，因为默认情况下，它不是基于线程安全的。此问题已通过tools / Makefile中的以下行修复：

 cd OpenBLAS; sed 's:# FCOMMON_OPT = -frecursive:FCOMMON_OPT =
                    -frecursive:' < Makefile.rule >tmp && mv tmp Makefile.rule

It's possible that the way ATLAS is built on Windows, by default it is not thread safe. However, you should be able to get around this problem by invoking
online2-wav-nnet2-latgen-faster with the option --num-threads-startup=1 which will limit it to 1 thread at startup. 在Windows上构建ATLAS的方式很有可能，默认情况下它不是线程安全的。--num-threads-startup \u003d 1将在启动时将其限制为1个线程。