search

MontrealCorpusTools / Montreal-Forced-Aligner

Command line utility for forced alignment using Kaldi
https://montrealcorpustools.github.io/Montreal-Forced-Aligner/
MIT License
1.26k stars 242 forks source link

Fatal error with Kaldi binaries #341

Open connieting opened 2 years ago

connieting commented 2 years ago

Hi MFA Team,

I'm trying to align a new dataset using 'mfa align' and am running into a Kaldi issue: When running, the training for mono runs fine and gets to 100%, but training for tri gives back this error: "WARNING - Subset specified is larger than the dataset, using full corpus for this training block. There was one or more errors when running Kaldi binaries. For more details, please check /[Directory]/train_and_align.log" Within the .log file, I'm not sure which portions are relevant for the issue, but the last lines are:

2021-09-28 15:18:27,458 - train_and_align - DEBUG - kaldi::KaldiFatalError

I've attached the end portion of the .log file as a picture for more info, in case helpful. (Running on Mac OS 11.6; MFA version 2.0.0b3)

Any suggestions to troubleshoot this would be much appreciated, thank you!

train_and_align
xiaofeng1020 commented 2 years ago

Hi，I encountered the same problem, did you solve it? Thank you！

mmcauliffe commented 2 years ago

Hi, can you try updating the latest version using the new conda build? https://montreal-forced-aligner.readthedocs.io/en/latest/getting_started.html#getting-started-ref I don't know for sure that it'll solve your issue, but they Kaldi binaries are now built on conda forge, so they should support more setups.