[BUG] Low Accuracy

[arnavmehta7]

Debugging checklist

[ x] Have you updated to latest MFA version? [ x] Have you tried rerunning the command with the --clean flag?

Describe the issue The accuracy for the timestamps is too low it is mostly off by 1 seconds on short audios and 6 seconds on long audios

For Reproducing your issue Download the original librispeech dataset and run the mfa align inputs lexicon.txt english_mfa outputs

1. Corpus structure

   • What language is the corpus in? English
   • How many files/speakers? 1
   • Are you using lab files or TextGrid files for input? Lab Files

2. Dictionary

   • Are you using a dictionary from MFA? If so, which one? lexicon
   • If it's a custom dictionary, what is the phoneset? NA

3. Acoustic model

   • If you're using an acoustic model, is it one download through MFA? If so, which one? english_mfa
   • If it's a model you've trained, what data was it trained on? NA

Log file Please attach the log file for the run that encountered an error (by default these will be stored in ~/Documents/MFA). No error

Desktop (please complete the following information):

• OS: [e.g. Windows, OSX, Linux] Linux
• Version [e.g. MacOSX 10.15, Ubuntu 20.04, Windows 10, etc] 20.04
• Any other details about the setup (Cloud, Docker, etc) WSL2

[mmcauliffe]

What does "lexicon.txt" look like? Can you share any example textgrids of the misalignments, or files that I can try to replicate and see what's going on?

[fangg2000]

Is there any way to improve accuracy? I found that the data is not accurate if the quantity is only a few. I have tried installing different versions, such as 1.0, 2.2 and 3.0, but in the 10 tests, the accuracy is not as good as the results I got on the win system before, and some of them are very different.

[mmcauliffe]

@fangg2000 can you create a new issue? I'm going to close this one as stale. If possible, please include the specifics of the runs you've done, there are multiple reasons why you're not seeing the same accuracy.