File size: 1,287 Bytes
2631d60
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
# @package __global__

classifier_free_guidance:
  training_dropout: 0.1
  inference_coef: 3.0

attribute_dropout:
  args:
    active_on_eval: false
  text: 
    description: 0.4
  wav:
    self_wav: 0.4

fuser:
  cross_attention_pos_emb: false
  cross_attention_pos_emb_scale: 1
  sum: []
  prepend: [self_wav, description]
  cross: []
  input_interpolate: []

conditioners:
  self_wav:
    model: style
    style:
      model_name: mert
      transformer_scale: default
      sample_rate: ${sample_rate}
      encodec_checkpoint: '//pretrained/facebook/encodec_32khz'
      encodec_n_q: 3
      length: 3.0
      ds_factor: 15 # Since MERT is 75Hz, 75/15 results into 5Hz representations
      n_q_out: 6
      eval_q: 3
      q_dropout: true
      bins: 1024
      varying_lengths: [1.5, 4.5]
      batch_norm: true
      compute_mask: true
      num_codebooks_lm: ${transformer_lm.n_q}
      ds_rate_compression: 640
      use_middle_of_segment: false
      rvq_threshold_ema_dead_code: 0.1

  description:
    model: t5
    t5:
      name: t5-base
      finetune: false
      word_dropout: 0.2
      normalize_text: false

dataset:
  train:
    merge_text_p: 0.25
    drop_desc_p: 0.5
    drop_other_p: 0.5
  shuffle: true