base:
  repo_id: SWivid/F5-TTS
  subfolder: F5TTS_v1_Base
  checkpoint: model_1250000.safetensors
  sha256: 670900fd14e6c458b95da6e9ed317cdb20dbaf7a1c02ac06a05475a9d32b6a38
  vocab: vocab.txt

adapter:
  repo_id: ScreamingTony/AffectF5
  filename: affectf5_c7_adapter.safetensors
  sha256: 26dd9fdc69395e8a937feadb3417e5503667eaedb6a85b12b26d0374a58f253f

model:
  dim: 1024
  depth: 22
  heads: 16
  ff_mult: 2
  text_dim: 512
  text_mask_padding: true
  qk_norm: null
  conv_layers: 4
  pe_attn_head: null
  attn_backend: torch
  attn_mask_enabled: false
  checkpoint_activations: false

control:
  control_dim: 64
  num_emotions: 6
  null_emotion_id: 0
  neutral_emotion_id: 1
  emotion_embedding_dim: 128
  emotion_hidden_dim: 256
  emotion_depth: 2
  arousal_fourier_dim: 64
  arousal_fourier_scale: 50.0
  arousal_fourier_max_period: 1000.0
  arousal_include_raw: true
  arousal_hidden_dim: 256
  arousal_depth: 2
  control_dropout: 0.0
  control_gate_init: 0.1
  control_zero_init: true
  injection:
    mask: target_only
    fusion: add
    strategy: axis_split_film_residual
    axis_split:
      emotion:
        film_layers: [11, 17]
        hidden_dim: 256
      arousal:
        residual_layers: [14, 20]
        hidden_dim: 256
        output_gain: 1.0
      time_gate_hidden_dim: 64

lora:
  targets: [to_q, to_k, to_v, to_out.0, ff.0.0, ff.2]
  rank: 16
  alpha: 32.0
  dropout: 0.05

inference:
  nfe_steps: 32
  global_guidance: 2.0
  emotion_guidance: 0.0
  arousal_guidance: 0.0
  arousal_scale: 0.75
  sway_sampling_coef: -1.0
  sample_rate: 24000
  hop_length: 256