yizhilll/MERT

Issues

• HuBERT and data2vec retrained on music

  #17 opened 4 months ago by lcc-404
  1

• Reproducing the numbers

  #13 opened a year ago by Laubeee
  1

• Music Descriptor prediction (especially for EMO task)

  #16 opened 6 months ago by uu95
  0

• Questions about data preparation

  #15 opened 8 months ago by nkkanee
  0

• will you provide fairseq checkpoints, so that tokenization from MERT is possible without retrain a clustering model?

  #4 opened 8 months ago by heimanba89
  1

• Fairseq Checkpoint

  #12 opened 8 months ago by lucasgris
  1

• When jointly mapping different modal features to the same semantics, which layer of the hidden layer is more appropriate to select

  #11 opened 8 months ago by tanggang1997
  1

• How to convert a fairseq checkpoint to a huggingface checkpoint

  #14 opened 8 months ago by z592694590
  1

• Which embedding to use for the downstream task of genre classification?

  #5 opened a year ago by elloza
  1

• questions about dictionary.encode_line to process encodec labels

  #8 opened a year ago by xujinchang
  6

• What is the sampling rate during training? Is it okay to upsample 16k to 24k during testing?

  #9 opened a year ago by xuanjihe
  1

• Regarding attention relaxation

  #10 opened a year ago by PetrosVav
  0

• Will the tokens be released in the future?

  #3 opened a year ago by A-Quarter-Mile
  1

• finetuned model for downstream tasks?

  #7 opened a year ago by jidanhuang
  1

• High vram usage for long audio.

  #6 opened a year ago by hykilpikonna
  1

• ModuleNotFoundError: No module named 'datasets'

  #2 opened 2 years ago by yygg678
  0