| { | |
| "model_type": "semacs", | |
| "semantic_hidden_size": 1280, | |
| "use_vocos": true, | |
| "model": { | |
| "codec_encoder": { | |
| "ngf": 48, | |
| "hidden_dim": 1280, | |
| "depth": 6, | |
| "heads": 12, | |
| "pos_meb_dim": 64, | |
| "up_ratios": [2, 2, 4, 4, 5], | |
| "dilations": [1, 3, 9], | |
| "output_dim": 1280 | |
| }, | |
| "codec_decoder": { | |
| "ngf": 48, | |
| "up_ratios": [5, 4, 4, 2, 2], | |
| "dilations": [1, 3, 9], | |
| "hidden_dim": 1024, | |
| "depth": 12, | |
| "heads": 16, | |
| "pos_meb_dim": 64, | |
| "vq_num_quantizers": 1, | |
| "vq_dim": 2048, | |
| "vq_commit_weight": 0.25, | |
| "vq_weight_init": false, | |
| "vq_full_commit_loss": false, | |
| "codebook_size": 16384, | |
| "codebook_dim": 16, | |
| "hop_length": 320, | |
| "sample_rate": 16000 | |
| } | |
| }, | |
| "architectures": [ | |
| "SEMACS" | |
| ] | |
| } | |