pause_model.json

{
  "format": "rubato-pause-model/1",
  "config": {
    "hidden": 64,
    "l2": 0.0001,
    "lr": 0.005,
    "epochs": 40,
    "batch_size": 4096,
    "seed": 0,
    "hazard_floor": 1e-05,
    "hazard_ceil": 0.75
  },
  "timing": {
    "frame_s": 0.02,
    "horizon_s": 3.0,
    "conflict_window_s": 2.0,
    "yield_window_s": 0.3,
    "counterfactual_slack_s": 1.5,
    "max_wait_s": 4.0,
    "min_pause_s": 0.18,
    "asr_lag_s": 0.25
  },
  "feature_names": [
    "time_basis[0]",
    "time_basis[1]",
    "time_basis[2]",
    "time_basis[3]",
    "time_basis[4]",
    "time_basis[5]",
    "time_basis[6]",
    "time_basis[7]",
    "time_basis[8]",
    "time_basis[9]",
    "time_basis[10]",
    "time_basis[11]",
    "turn:prev_dur_log",
    "turn:holds_so_far",
    "asr:compl_heuristic",
    "asr:compl_learned",
    "asr:ends_filler",
    "asr:ends_conjunction",
    "asr:ends_function",
    "asr:ends_backchannel",
    "asr:wh_initial",
    "asr:aux_initial",
    "asr:n_tokens_log",
    "asr:final_lengthening",
    "asr:trailing_repeat",
    "asr:filler_density",
    "asr:prev_rate",
    "asr:prev_rate_z",
    "asr:internal_pause_rate",
    "prosody:f0_slope_st_s",
    "prosody:f0_end_z",
    "prosody:f0_range_st",
    "prosody:energy_slope_db_s",
    "prosody:energy_end_db",
    "prosody:voiced_frac",
    "speaker:spk_hold_rate",
    "speaker:spk_gap_log_mean",
    "speaker:spk_gap_log_sd",
    "speaker:spk_rate_ema",
    "speaker:spk_pause_log_mean",
    "speaker:spk_n_obs_log",
    "dyn:residual_energy_db",
    "dyn:residual_energy_slope",
    "dyn:breath_active",
    "dyn:breath_recency",
    "mask:asr",
    "mask:acoustic"
  ],
  "feature_dim": 47,
  "train_report": {
    "val_nll": 0.024722312995084168,
    "val_nll_calibrated": 0.02471667232808826,
    "train_nll": 0.026774539329211252,
    "base_rate": 0.005187452746528725,
    "n_frames": 1071851.0
  }
}