raratu's picture
Add verified Irodori-TTS-v4.1-Small iOS ONNX pack
d9fe87b verified
Raw History Blame Contribute Delete
2.07 kB
{
"model": {
"latent_dim": 32,
"latent_patch_size": 1,
"model_dim": 1280,
"num_layers": 12,
"num_heads": 20,
"mlp_ratio": 2.875,
"text_mlp_ratio": 2.6,
"speaker_mlp_ratio": 2.6,
"dropout": 0.0,
"text_vocab_size": 102400,
"text_tokenizer_repo": "sbintuitions/modernbert-ja-310m",
"text_encoder_revision": "77675fc96a7e445e982e2ba90246b816efc74ec6",
"text_add_bos": true,
"text_encoder_type": "pretrained",
"pretrained_projector_type": "residual_mlp",
"pretrained_projector_hidden_ratio": 2.0,
"pretrained_projector_dropout": 0.0,
"text_dim": 512,
"text_layers": 14,
"text_heads": 10,
"use_caption_condition": true,
"use_speaker_condition": true,
"caption_vocab_size": 102400,
"caption_tokenizer_repo": "sbintuitions/modernbert-ja-310m",
"caption_add_bos": true,
"caption_dim": 512,
"caption_layers": null,
"caption_heads": null,
"caption_mlp_ratio": null,
"speaker_dim": 768,
"speaker_layers": 8,
"speaker_heads": 12,
"speaker_patch_size": 4,
"timestep_embed_dim": 512,
"adaln_rank": 192,
"norm_eps": 1e-05,
"use_duration_predictor": true,
"duration_aux_dim": 14,
"duration_hidden_dim": 1024,
"duration_layers": 3,
"duration_dropout": 0.1,
"duration_attention_heads": 8,
"duration_architecture": "token_sum_dual_adarn_zero_no_aux",
"duration_token_init_frames": 9.0,
"duration_speaker_fusion": "adarn_zero",
"duration_caption_fusion": "adarn_zero",
"duration_caption_pooling": "masked_mean"
},
"inference": {
"max_text_len": 256,
"max_caption_len": 512,
"ref_max_seconds": 120.0
},
"export": {
"seq_len": 128,
"text_len": 256,
"caption_len": 512,
"ref_len": 200,
"sample_rate": 48000,
"hop_length": 1920,
"requires_duration": true,
"shared_condition_encoder": true,
"context_kv_cache": true,
"latent_dim": 32,
"num_layers": 12,
"has_caption": true,
"has_speaker_kv_scale": true,
"lora_merged": false
}
}