| { | |
| "encoder": { | |
| "n_mels": 80, | |
| "d_model": 512, | |
| "n_heads": 8, | |
| "n_layers": 20, | |
| "ff_dim": 2048, | |
| "conv_channels": [ | |
| 512, | |
| 512 | |
| ], | |
| "conv_kernel": 3, | |
| "conv_stride": [ | |
| 1, | |
| 1 | |
| ], | |
| "pos_encoding": "rope", | |
| "norm": "rms", | |
| "activation": "gelu", | |
| "dropout": 0.1, | |
| "bidirectional": true, | |
| "layer_scale_init": 0.05, | |
| "rope_learnable": true, | |
| "qk_norm": true | |
| }, | |
| "decoder": { | |
| "d_model": 512, | |
| "n_heads": 8, | |
| "n_layers": 20, | |
| "ff_dim": 2048, | |
| "n_codebooks": 9, | |
| "vocab_size": 1024, | |
| "pad_token_id": 1024, | |
| "bos_token_id": 1024, | |
| "pos_encoding": "rope", | |
| "norm": "rms", | |
| "activation": "gelu", | |
| "dropout": 0.1, | |
| "cross_attn_every_layer": true, | |
| "rope_learnable": true, | |
| "qk_norm": true, | |
| "layer_scale_init": 0.05 | |
| }, | |
| "depth": { | |
| "d_model": 384, | |
| "n_heads": 6, | |
| "n_layers": 4, | |
| "ff_dim": 1536, | |
| "dropout": 0.1, | |
| "qk_norm": true, | |
| "layer_scale_init": 0.05 | |
| }, | |
| "mel": { | |
| "sample_rate": 24000, | |
| "n_fft": 1024, | |
| "hop_length": 280, | |
| "win_length": 1024, | |
| "n_mels": 80, | |
| "fmin": 0, | |
| "fmax": 8000, | |
| "log_clamp_db": -80, | |
| "normalization": "minmax" | |
| }, | |
| "dac": { | |
| "model": "44khz", | |
| "sample_rate": 44100, | |
| "n_codebooks": 9, | |
| "vocab_size": 1024, | |
| "frame_rate": 86.13, | |
| "checkpoint_path": null | |
| } | |
| } |