ms_clap/src/configs/config_2023.yml

# TEXT ENCODER CONFIG
text_model: 'gpt2'
text_len: 77
transformer_embed_dim: 768
freeze_text_encoder_weights: True

# AUDIO ENCODER CONFIG
audioenc_name: 'HTSAT'
out_emb: 768
sampling_rate: 44100
duration: 7
fmin: 50
fmax: 8000 #14000 
n_fft: 1024 # 1028 
hop_size: 320
mel_bins: 64
window_size: 1024

# PROJECTION SPACE CONFIG 
d_proj: 1024
temperature: 0.003

# TRAINING AND EVALUATION CONFIG
num_classes: 527
batch_size: 1024
demo: False