kaldi-asr/kaldi

Update single job scripts like utils/data/get_utt2dur.sh, steps/compute_cmvn_stats.sh to use multiple jobs

Open

#984 创建于 2016年8月20日

在 GitHub 查看
 (6 评论) (0 反应) (0 负责人)Shell (15,392 star) (5,359 fork)batch import
enhancementhelp wantedstale

描述

In the large data recipes in Kaldi many single job scripts in steps/* and utils/*, which were very fast for normal sized LVCSR tasks, consume several hours. e.g. on fisher_swbd/aspire the utils/data/get_utt2dur.sh script runs for a few hours.

We need to update these scripts to add parallelization support.

Required skill: shell scripting

贡献者指南

技术栈
shell
领域
dataperformance
议题类型
performance
难度面向新贡献者的预计实现难度，1 表示很小改动，5 表示专家级工作。
3
预计时间有经验贡献者完成调查、实现、测试并准备 pull request 的粗略时间范围。
1-2 days
活动状态议题当前的可参与程度：新鲜、活跃、陈旧、阻塞或等待维护者输入。
stale
清晰度议题是否清楚说明期望改动、验收标准和下一步。
clear
前置要求
shell scriptingparallelization experience
新手友好度1-100 的估计分数，表示该议题对首次贡献者的友好程度。
45
研究方向
First, examine existing single job scripts like utils/data/get utt2dur.sh and steps/compute cmvn stats.sh to understand their current structure. Then, look at how Kaldi already implements parallel jobs in other scripts (e.g., using utils/run.pl or utils/parallel.pl). The goal is to modify these scripts to split work across multiple jobs while maintaining correctness. Finally, test with large datasets to verify performance improvement and ensure no regression. Since the issue is from 2016, check if any partial work exists in the repository or related forks.