{ "metadata_schema": "streaming_pcn_gopt_v2_stateful", "pcn_type": "nbest_derived_phone_confusion_network", "phone_dim": 40, "seq_len": 72, "prosody_dim": 14, "prosody_norm_mean": [ 3.9681200981140137, 204.3585205078125, 24.05475616455078, -0.20867417752742767, 0.12271453440189362, -4.253726005554199, 1.724206805229187, 0.004727904684841633, 0.20122256875038147, 10.172670364379883, 0.3134472370147705, 1.937715768814087, 5.8537139892578125, 17295600.0 ], "prosody_norm_std": [ 2.0985183715820312, 55.536617279052734, 13.517866134643555, 0.4277317225933075, 0.06661950051784515, 0.6672922372817993, 0.378594309091568, 0.009153330698609352, 0.0010623399866744876, 11.080301284790039, 0.1449834257364273, 0.9631871581077576, 2.870692014694214, 205320480.0 ], "slot_prosody_dim": 17, "slot_prosody": [ "slot_duration", "slot_log_energy_mean", "slot_log_energy_std", "slot_log_energy_max", "slot_f0_mean", "slot_f0_std", "slot_f0_max", "slot_voiced_ratio", "slot_position_in_word", "word_phone_count", "energy_relative_to_word_mean", "duration_relative_to_word_mean", "f0_relative_to_word_mean", "is_vowel", "lexical_stress_0", "lexical_stress_1", "lexical_stress_2" ], "slot_prosody_norm_mean": [ 0.039656512439250946, -0.9386982917785645, 0.12416643649339676, -0.7695538997650146, 49.076961517333984, 1.3523738384246826, 51.32339096069336, 0.18721239268779755, 0.13828140497207642, 1.0300568342208862, 4.231509587171445e-10, 0.28040772676467896, -2.2446879910376083e-08, 0.11200845241546631, 0.0, 0.0, 0.0 ], "slot_prosody_norm_std": [ 0.11444349586963654, 1.6153852939605713, 0.31157857179641724, 1.3436129093170166, 92.65617370605469, 4.188445568084717, 97.01981353759766, 0.3661966025829315, 0.31229057908058167, 1.9228776693344116, 0.3923450708389282, 0.5529012084007263, 27.52008056640625, 0.31510770320892334, 0.0, 0.0, 0.0 ], "utt_pooling_head": "gru_visible", "fusion_mode": "scalar_gate", "stress_branch": "detached", "stress_grad_scale": 0.2, "uses_state_projection": false, "args": { "lr": 0.001, "n_epochs": 80, "batch_size": 12, "num_workers": 4, "embed_dim": 64, "depth": 3, "heads": 4, "gru_dim": 64, "main_context_tokens": 16, "tbptt_steps": 0, "loss_w_phone": 1.0, "loss_w_word": 1.0, "loss_w_utt": 1.0, "loss_w_asr": 0.5, "loss_w_uncertainty": 0.2, "loss_w_confidence": 0.2, "loss_w_abstention": 0.2, "loss_w_calibration": 0.1, "loss_w_teacher_score": 0.8, "loss_w_prefix_kd": 1.0, "loss_w_rank": 0.2, "loss_w_oracle_phone": 0.3, "loss_w_oracle_word": 1.0, "loss_w_oracle_utt_prefix": 0.2, "loss_w_oracle_utt_final": 0.3, "loss_w_stress_pearson": 0.2, "loss_w_oracle_stress_pearson": 0.3, "loss_w_teacher_stress_pearson": 0.1, "loss_w_stress_rank": 0.1, "loss_w_oracle_stress_rank": 0.1, "loss_w_phone_stability": 0.01, "loss_w_word_stability": 0.01, "loss_w_utt_stability": 0.01, "loss_w_commit_consistency": 0.0, "loss_w_state_projection": 0.0, "utt_dim_weights": [ 1.0, 0.0, 1.0, 1.0, 1.0 ], "word_dim_weights": [ 1.0, 3.0, 1.0 ], "teacher_word_dim_weights": [ 1.0, 2.0, 1.0 ], "oracle_word_dim_weights": [ 1.0, 5.0, 1.0 ], "soft_label_policy": "original", "relaxed_min_gt_posterior": 0.05, "relaxed_acoustic_scale": 0.3, "relaxed_min_weight": 0.05, "relaxed_max_weight": 0.5, "utt_pooling_head": "gru_visible", "fusion_mode": "scalar_gate", "rank_margin": 0.02, "stress_rank_margin": 0.02, "stress_rank_max_items": 512, "stress_loss_mask": "vowel", "stress_voiced_threshold": 0.3, "stress_branch": "detached", "stress_grad_scale": 0.2, "grad_clip_norm": 1.0, "compile": false, "tf32": true, "seed": 1337 }, "preserve_unused_legacy_utt_head": true }