configs/audio2motion/model/audio_processer_config.yaml

# models settings
model_params:
  model_name: hubert   # wav2vec or hubert
  model_type: base   # base large
  is_chinese: True
  is_original: True
  only_last_features: False
  use_audio_separator: False
  audio_separator_name: Kim_Vocal_2.onnx

# model weights
model_weights:
  audio_separator_path: pretrain_weights/audio/audio_separator
  hubert_path:
    chinese:
      base: pretrain_weights/audio/chinese-hubert-base
# data settings
data_params:
  sample_rate: 16000
  max_length: 60         # seconds
  sub_clip_length: 3000  # samples
  fps: 25
  sample_strategy: "presample"
  audio_pad_mode: replicate  # pad mode for audio, replicate or zero
  save_to_cpu: True    # saving gpu memory

# device settings
device_params:
  device_id: 0
  flag_force_cpu: False
  flag_use_half_precision: False
  
cache_dir: preprocessed/HDTF/vocals
tmp_dir: src/tmp