web/voices/italian/meta.json
| 1 | { |
| 2 | "key": "italian", |
| 3 | "espeak_voice": "it", |
| 4 | "g2p_voice_slot": 12, |
| 5 | "length_scale": 1.0, |
| 6 | "sample_rate": 22050, |
| 7 | "front_bytes": 1670094, |
| 8 | "dec_bytes": 1461858, |
| 9 | "front_dims": { |
| 10 | "d_vocab": 143, |
| 11 | "d_hidden": 64, |
| 12 | "d_depth": 3, |
| 13 | "d_kernel": 5, |
| 14 | "d_max_tokens": 1067, |
| 15 | "d_max_duration": 80, |
| 16 | "a_vocab": 143, |
| 17 | "a_hidden": 96, |
| 18 | "a_token_depth": 3, |
| 19 | "a_depth": 4, |
| 20 | "a_kernel": 5, |
| 21 | "a_out_channels": 192, |
| 22 | "adapter_mode": 0, |
| 23 | "adapter_kernel": 0, |
| 24 | "adapter_rank": 0, |
| 25 | "weight_floats": 834763, |
| 26 | "meta_bytes": 568, |
| 27 | "n_tensors": 62 |
| 28 | }, |
| 29 | "dec_dims": { |
| 30 | "in_ch": 192, |
| 31 | "c0": 160, |
| 32 | "c1": 80, |
| 33 | "c2": 40, |
| 34 | "c3": 20, |
| 35 | "pf_channels": 0, |
| 36 | "pf_layers": 0, |
| 37 | "pf_kernel": 9, |
| 38 | "pf_scale": 0.25, |
| 39 | "hop": 256, |
| 40 | "weight_floats": 730721, |
| 41 | "meta_bytes": 416, |
| 42 | "n_tensors": 46 |
| 43 | }, |
| 44 | "checkpoints": { |
| 45 | "duration": "artifacts/voices/it_IT/duration/duration-student.pt", |
| 46 | "acoustic": "artifacts/voices/it_IT/joint/latent-student.pt", |
| 47 | "decoder": "artifacts/voices/it_IT/joint/decoder-student.pt" |
| 48 | }, |
| 49 | "front": "front_f16.bin", |
| 50 | "front_widened_sha256": "f6acf50b09a5d314a1066ae3ad2b25dd1fe5dc64b3086c70a91040d2b504009f", |
| 51 | "dec": "dec_f16.bin", |
| 52 | "dec_widened_sha256": "529e859d5cc285152253a560f9e34f4d1fbf1b7e52fdbb6b5c3288d6a3c7f288", |
| 53 | "weights": "f16", |
| 54 | "weights_note": "float16 storage, widened to float32 by the loader before it reaches the runtime. float16 is the reference for this voice: it is the precision of the roota.raw-fp16.v1 package every CER/WER number was measured on." |
| 55 | } |
| 56 | |