What?

Fixed bugs when running egs2/voxtlm_v1/lm1/run.sh. Specifically:

• Add missing files (commit c69c0a0, ddad665)
• Fix path related problems (commit 0dffa60, 08f322f, f5f0121)
• Fix variable related problems (commit ddad665, ed37105, 4f920f4, 22cc663, d6915e8)
• Fix typo (commit 10e6ec6)

Notes

• Here I kept the changes minimal so that current recipe runs smoothly. More changes could be made to improve the code.
  • Especially I think it would be better to include TTS unit-to-wav generation stage in the recipe. Help from the original authors is appreciated.
• I ran the code with reduced data to speed up the debugging process. I did not check the performance of full training.

Why?

The recipe did not run smoothly as it is. See the issue below for more detail.

See also

• Bugs in reproducing VoxtLM v1 #5777