models:
  inpainter: checkpoint_step_16799_loss_5.342.pkl
  text_tokenizer: tokenizer-multi_bpe16384_merged_extended_58M_nonverbal.json
  speech_tokenizer: xlsr2_1b_v2_custom.pt
  kmeans_layer: kmeans_10k.npy
  voice_encoder: voice_encoder_1992000.pt
  vocoder: v090_g_01105000