| { | |
| "semantic_model": "facebook/w2v-bert-2.0", | |
| "semantic_layer_index": 16, | |
| "sr": 16000, | |
| "output_sr": 44100, | |
| "codec_encoder": { | |
| "ngf": 48, | |
| "hidden_dim": 1024, | |
| "up_ratios": [ | |
| 2, | |
| 2, | |
| 4, | |
| 4, | |
| 5 | |
| ], | |
| "dilations": [ | |
| 1, | |
| 3, | |
| 9 | |
| ], | |
| "activation": "snake_beta" | |
| }, | |
| "codec_decoder": { | |
| "hidden_dim": 1024, | |
| "activation": "snake_beta", | |
| "conv_padding_mode": "causal_full", | |
| "resnet_kernel_size": 1, | |
| "embed_kernel_size": 3, | |
| "depth": 12, | |
| "heads": 16, | |
| "pos_emb_dim": 64, | |
| "vq_dim": 2048, | |
| "attn_causal": true, | |
| "mlp_expansion": 4, | |
| "upsampler_factors": [ | |
| 3 | |
| ], | |
| "upsampler_kernel_sizes": [ | |
| 9 | |
| ], | |
| "output_n_fft": 1176, | |
| "output_hop_length": 294, | |
| "quantizer": { | |
| "mode": "rvq", | |
| "num_quantizers": 16, | |
| "codebook_size": 4096, | |
| "fsq_levels": [ | |
| 8, | |
| 8, | |
| 8, | |
| 8 | |
| ], | |
| "cb1_type": "fsq", | |
| "residual_type": "vq", | |
| "vq_codebook_dim": 8 | |
| } | |
| }, | |
| "firing": { | |
| "enabled": true, | |
| "mode": "adaptive", | |
| "target_ratio": 0.25, | |
| "max_gap": 64, | |
| "decision": "clock", | |
| "hnet_ms_offsets": [ | |
| [ | |
| 1, | |
| 0 | |
| ], | |
| [ | |
| 1, | |
| 1 | |
| ], | |
| [ | |
| 2, | |
| 1 | |
| ] | |
| ], | |
| "head_in_budget": true, | |
| "aggregate": "mean", | |
| "signals": [ | |
| "hnet_ms", | |
| "flux" | |
| ] | |
| }, | |
| "inference": { | |
| "attention_window_seconds": 6.0, | |
| "decode_window_frames": 300, | |
| "decode_stride_frames": 100, | |
| "decode_crossfade_ms": 10.0 | |
| } | |
| } | |