{
  "model_type": "f5_tts_dit_v1_base",
  "architecture": "F5-TTS (DiT variant)",
  "dim": 1024,
  "depth": 22,
  "heads": 16,
  "ff_mult": 2,
  "text_dim": 512,
  "conv_layers": 4,
  "n_params_approx": "335M",
  "mel_spec": {
    "n_mels": 100,
    "n_fft": 1024,
    "hop_length": 256,
    "win_length": 1024,
    "sample_rate": 24000
  },
  "vocoder": "charactr/vocos-mel-24khz",
  "tokenizer": "char",
  "vocab_size": 2704
}