{ "sample_rate": 16000, "compute_eval_loss": false, "log_prediction": true, "rnnt_reduction": "mean_volume", "skip_nan_grad": true, "model_defaults": { "enc_hidden": 512, "pred_hidden": 640, "joint_hidden": 640, "tdt_durations": [ 0, 1, 2, 3, 4 ], "num_tdt_durations": 5 }, "train_ds": { "manifest_filepath": null, "sample_rate": 16000, "batch_size": null, "shuffle": true, "num_workers": 8, "pin_memory": true, "max_duration": 40, "min_duration": 0.1, "is_tarred": true, "tarred_audio_filepaths": null, "shuffle_n": 2048, "bucketing_strategy": "fully_randomized", "bucketing_batch_size": null, "shard_manifests": true, "use_lhotse": true, "use_bucketing": true, "num_buckets": 30, "bucket_duration_bins": null, "batch_duration": 600, "defer_setup": true }, "validation_ds": { "manifest_filepath": null, "sample_rate": 16000, "batch_size": 32, "shuffle": false, "use_start_end_token": false, "num_workers": 8, "pin_memory": true }, "test_ds": null, "tokenizer": { "dir": "/tokenizers/en_asrset_PnC_suno_PnC_tokenizer/tokenizer_spe_bpe_v1024/", "type": "bpe", "model_path": "nemo:a4c254472f6a4c388105ae0109636cdc_tokenizer.model", "vocab_path": "nemo:dad8b72697f34006bfd597c3a560ea75_vocab.txt", "spe_tokenizer_vocab": "nemo:0f176c5829c3472daa16f5065b15081d_tokenizer.vocab" }, "preprocessor": { "_target_": "nemo.collections.asr.modules.AudioToMelSpectrogramPreprocessor", "sample_rate": 16000, "normalize": "per_feature", "window_size": 0.025, "window_stride": 0.01, "window": "hann", "features": 80, "n_fft": 512, "frame_splicing": 1, "dither": 1e-05, "pad_to": 0 }, "spec_augment": { "_target_": "nemo.collections.asr.modules.SpectrogramAugmentation", "freq_masks": 2, "time_masks": 10, "freq_width": 27, "time_width": 0.05 }, "encoder": { "_target_": "nemo.collections.asr.modules.ConformerEncoder", "feat_in": 80, "feat_out": -1, "n_layers": 17, "d_model": 512, "subsampling": "dw_striding", "subsampling_factor": 8, "subsampling_conv_channels": 256, "causal_downsampling": false, "reduction": null, "reduction_position": null, "reduction_factor": 1, "ff_expansion_factor": 4, "self_attention_model": "rel_pos", "n_heads": 8, "att_context_size": [ -1, -1 ], "att_context_style": "regular", "xscaling": false, "untie_biases": true, "pos_emb_max_len": 5000, "conv_kernel_size": 9, "conv_norm_type": "batch_norm", "conv_context_size": null, "dropout": 0.1, "dropout_pre_encoder": 0.1, "dropout_emb": 0.0, "dropout_att": 0.1, "stochastic_depth_drop_prob": 0.0, "stochastic_depth_mode": "linear", "stochastic_depth_start_layer": 1 }, "decoder": { "_target_": "nemo.collections.asr.modules.RNNTDecoder", "normalization_mode": null, "random_state_sampling": false, "blank_as_pad": true, "prednet": { "pred_hidden": 640, "pred_rnn_layers": 1, "t_max": null, "dropout": 0.2 }, "vocab_size": 1024 }, "joint": { "_target_": "nemo.collections.asr.modules.RNNTJoint", "log_softmax": null, "preserve_memory": false, "fuse_loss_wer": true, "fused_batch_size": 4, "jointnet": { "joint_hidden": 640, "activation": "relu", "dropout": 0.2, "encoder_hidden": 512, "pred_hidden": 640 }, "num_extra_outputs": 5, "num_classes": 1024, "vocabulary": [ "", "\u2581t", "\u2581th", "\u2581a", "in", "re", "\u2581the", "\u2581w", "\u2581s", "\u2581o", "er", "ou", "at", "nd", "it", "\u2581h", "\u2581c", "\u2581b", "is", "en", "on", "ing", "\u2581f", "\u2581to", "\u2581m", "es", "\u2581p", "or", "an", "\u2581d", "ll", "\u2581I", "ed", "\u2581and", "\u2581l", "\u2581of", "\u2581in", "\u2581y", "ar", "\u2581g", "\u2581you", "as", "om", "\u2581n", "ve", "\u2581that", "le", "ic", "us", "ow", "et", "al", "\u2581e", "ut", "\u2581it", "ot", "\u2581be", "\u2581T", "ion", "\u2581is", "\u2581wh", "\u2581re", "\u2581on", "\u2581we", "ent", "\u2581A", "ay", "\u2581ha", "\u2581Th", "id", "\u2581S", "ac", "gh", "ver", "ke", "\u2581for", "im", "ly", "ur", "ld", "\u2581he", "\u2581st", "all", "ro", "st", "se", "ct", "ith", "ir", "am", "\u2581this", "if", "\u2581W", "oo", "ri", "\u2581was", "ght", "\u2581u", "\u2581with", "ad", "ch", "\u2581se", "\u2581k", "\u2581an", "\u2581The", "\u2581li", "\u2581do", "\u2581B", "\u2581have", "\u2581as", "th", "\u2581are", "\u2581sh", "ust", "ce", "ally", "ill", "\u2581H", "\u2581j", "ter", "\u2581go", "\u2581And", "ation", "\u2581C", "\u2581so", "ome", "\u2581not", "op", "il", "ore", "\u2581ne", "\u2581can", "\u2581me", "\u2581at", "ould", "ant", "\u2581M", "\u2581like", "ere", "\u2581they", "ra", "ers", "\u2581ab", "\u2581de", "\u2581kn", "ge", "\u2581Y", "\u2581ch", "ul", "pp", "\u2581or", "\u2581al", "\u2581con", "\u2581com", "ess", "\u2581su", "out", "\u2581your", "\u2581So", "ate", "\u2581one", "\u2581all", "\u2581ex", "est", "\u2581fr", "\u2581just", "\u2581pro", "\u2581know", "\u2581O", "ain", "\u2581but", "ol", "ive", "\u2581v", "use", "very", "art", "qu", "\u2581my", "el", "\u2581N", "nt", "\u2581It", "\u2581what", "ab", "\u2581P", "\u2581wor", "\u2581out", "\u2581there", "\u2581up", "um", "\u2581from", "pe", "\u2581tw", "\u2581r", "and", "ight", "ort", "un", "\u2581L", "ist", "\u2581about", "ide", "ig", "ake", "\u2581D", "em", "os", "king", "rou", "ind", "our", "res", "\u2581We", "\u2581get", "\u2581E", "\u2581G", "ack", "\u2581le", "ity", "od", "\u2581F", "ard", "\u2581pl", "\u2581our", "\u2581int", "ment", "\u2581will", "ies", "\u2581by", "ink", "ca", "\u2581if", "red", "her", "ie", "\u2581us", "\u2581some", "\u2581don", "ven", "ood", "ast", "\u2581R", "\u2581his", "\u2581tim", "\u2581tr", "\u2581more", "ich", "ous", "ame", "\u2581going", "\u2581had", "\u2581them", "ook", "\u2581pe", "\u2581Wh", "\u2581You", "\u2581But", "ine", "\u2581here", "\u2581would", "cause", "right", "so", "ost", "ure", "\u2581has", "ect", "\u2581think", "\u2581fe", "ong", "\u2581see", "\u2581when", "\u2581who", "\u2581were", "\u2581really", "\u2581their", "\u2581want", "one", "ople", "\u2581then", "\u2581time", "\u2581sa", "ap", "\u2581te", "\u2581He", "\u2581ye", "ck", "\u2581her", "\u2581thing", "\u2581right", "\u2581which", "itt", "ice", "act", "\u2581people", "ty", "\u2581two", "\u2581J", "\u2581im", "ther", "ci", "ose", "\u2581cl", "\u2581qu", "\u2581man", "\u2581also", "ree", "\u2581en", "ud", "\u2581how", "reat", "ak", "hing", "ag", "\u2581any", "ff", "ace", "per", "\u2581because", "\u2581very", "own", "\u2581ad", "\u2581act", "\u2581been", "\u2581now", "\u2581ag", "\u2581into", "\u2581comp", "ars", "ions", "are", "ite", "iv", "\u2581these", "ays", "ep", "\u2581This", "\u2581she", "ans", "ah", "een", "\u2581over", "ry", "\u2581lo", "age", "\u2581pr", "\u2581sp", "ue", "\u2581co", "ick", "ber", "\u2581did", "ip", "ach", "\u2581back", "\u2581no", "\u2581cont", "\u2581other", "\u2581every", "pt", "\u2581need", "\u2581him", "\u2581U", "\u2581In", "\u2581work", "irst", "\u2581part", "\u2581look", "ittle", "ble", "iz", "\u2581un", "\u2581make", "omet", "nder", "ish", "na", "\u2581little", "\u2581off", "\u2581than", "\u2581got", "ually", "\u2581per", "\u2581good", "\u2581way", "\u2581could", "\u2581ac", "\u2581imp", "able", "\u2581where", "iff", "\u2581That", "\u2581res", "ount", "pl", "ance", "\u2581first", "\u2581ro", "\u2581pre", "ass", "\u2581say", "int", "ated", "ire", "uch", "ase", "\u2581somet", "ound", "\u2581down", "\u2581diff", "sel", "\u2581gu", "\u2581am", "ress", "\u2581lot", "ence", "\u2581dis", "orm", "ix", "\u2581po", "ving", "enty", "\u2581K", "\u2581spe", "und", "he", "\u2581much", "\u2581ar", "round", "\u2581app", "co", "ark", "\u2581new", "ater", "ult", "end", "\u2581even", "\u2581start", "ations", "rough", "ile", "fter", "\u2581well", "be", "\u2581They", "\u2581three", "ign", "ild", "\u2581said", "ough", "ang", "\u2581too", "ade", "\u2581bl", "ens", "\u2581inc", "ia", "\u2581those", "\u2581mo", "\u2581take", "\u2581through", "\u2581fl", "\u2581kind", "\u2581things", "\u2581bet", "\u2581only", "\u2581St", "\u2581let", "cess", "\u2581Ch", "ary", "vel", "\u2581If", "xt", "other", "av", "ical", "ord", "\u2581again", "\u2581something", "onna", "fore", "\u2581may", "ting", "\u2581bu", "\u2581differe", "urn", "\u2581gonna", "\u2581does", "uct", "og", "\u2581twenty", "\u2581gr", "\u2581Ye", "wn", "\u2581should", "\u2581comm", "ition", "\u2581under", "\u2581hel", "ory", "\u2581fo", "\u2581use", "igh", "ife", "\u2581actually", "\u2581tal", "\u2581call", "ents", "ious", "ull", "\u2581There", "\u2581Yeah", "\u2581most", "\u2581ke", "ors", "ved", "ys", "\u2581sc", "\u2581happ", "ope", "\u2581help", "atch", "\u2581What", "\u2581rem", "ple", "\u2581Now", "\u2581br", "ool", "oth", "\u2581four", "self", "\u2581str", "ne", "thing", "\u2581put", "ial", "\u2581great", "ail", "ub", "ning", "\u2581sm", "\u2581feel", "\u2581five", "ody", "undred", "iss", "ank", "get", "aking", "\u2581many", "\u2581hundred", "\u2581years", "\u2581being", "\u2581come", "\u2581mean", "ily", "\u2581different", "\u2581after", "\u2581ser", "\u2581show", "form", "ful", "oy", "\u2581six", "\u2581vide", "\u2581V", "\u2581its", "\u2581point", "\u2581day", "\u2581des", "ons", "\u2581bit", "\u2581bel", "\u2581before", "\u2581aw", "\u2581end", "\u2581Oh", "\u2581still", "ath", "\u2581long", "\u2581'", "ise", "ob", "day", "\u2581add", "ft", "ves", "ces", "ady", "\u2581cr", "\u2581around", "\u2581try", "les", "vers", "kay", "ian", "ates", "\u2581find", "ward", "\u2581As", "\u2581eight", "lic", "\u2581same", "\u2581pos", "\u2581em", "\u2581made", "\u2581supp", "\u2581life", "\u2581Be", "pect", "\u2581dec", "\u2581play", "ange", "\u2581att", "\u2581pers", "ways", "\u2581high", "\u2581hand", "\u2581next", "\u2581cons", "\u2581own", "\u2581inv", "ower", "\u2581ind", "ert", "ng", "ave", "\u2581year", "\u2581big", "ating", "\u2581world", "\u2581rel", "\u2581sure", "\u2581tra", "ew", "ered", "\u2581fin", "\u2581Well", "\u2581sl", "\u2581doing", "bs", "\u2581set", "\u2581rec", "ual", "cial", "\u2581ph", "erm", "\u2581love", "ph", "\u2581real", "\u2581last", "ict", "\u2581bo", "\u2581ra", "ible", "\u2581wr", "mer", "\u2581count", "ities", "\u2581always", "inet", "ments", "uc", "\u2581might", "\u2581inter", "\u2581video", "gin", "\u2581tell", "\u2581never", "vent", "\u2581import", "ied", "\u2581sy", "\u2581How", "ically", "ought", "\u2581thir", "\u2581rep", "ks", "ib", "\u2581fam", "ject", "\u2581bas", "\u2581She", "\u2581give", "akes", "\u2581ninet", "\u2581reg", "\u2581min", "\u2581op", "\u2581def", "\u2581didn", "te", "\u2581cour", "\u2581why", "\u2581ent", "\u2581place", "\u2581ins", "\u2581car", "ather", "\u2581person", "ular", "\u2581inst", "\u2581prod", "lect", "\u2581Al", "\u2581today", "\u2581bec", "\u2581sur", "\u2581All", "\u2581another", "\u2581bus", "\u2581keep", "ell", "ese", "riend", "\u2581quest", "\u2581talk", "als", "ings", "\u2581mon", "cond", "old", "\u2581acc", "\u2581la", "\u2581num", "ident", "\u2581che", "iness", "\u2581turn", "\u2581ear", "\u2581No", "ousand", "\u2581better", "ific", "\u2581loo", "\u2581gl", "oc", "\u2581important", "ited", "\u2581An", "\u2581thousand", "ility", "llow", "\u2581used", "\u2581gen", "\u2581sim", "li", "\u2581happen", "\u2581Un", "\u2581Let", "air", "ock", "ably", "gg", "\u2581watch", "\u2581For", "\u2581sw", "ren", "ute", "ever", "\u2581pol", "\u2581sch", "\u2581When", "\u2581such", "\u2581fif", "\u2581home", "\u2581cle", "\u2581contin", "ouse", "\u2581friend", "uring", "\u2581Okay", "gr", "\u2581able", "\u2581stud", "\u2581eff", "hip", "body", "\u2581top", "ness", "\u2581exper", "\u2581pret", "\u2581both", "\u2581done", "cri", "\u2581mark", "\u2581while", "\u2581old", "ros", "ont", "\u2581second", "ative", "\u2581thought", "\u2581best", "\u2581found", "iew", "\u2581belie", "\u2581each", "erest", "\u2581tri", "\u2581eas", "\u2581ca", "\u2581fact", "\u2581care", "\u2581fun", "atter", "ures", "\u2581head", "\u2581lear", "\u2581water", "\u2581hard", "\u2581few", "\u2581side", "ween", "\u2581exp", "\u2581away", "its", "\u2581ext", "lud", "\u2581run", "\u2581trans", "ince", "\u2581sk", "\u2581open", "cus", "\u2581between", "\u2581called", "\u2581wee", "\u2581pretty", "ason", "\u2581far", "ember", "omm", "\u2581interest", "any", "ner", "uff", "\u2581pres", "\u2581cur", "\u2581child", "ee", "\u2581toget", "\u2581together", "olog", "\u2581God", "ond", "\u2581char", "\u2581looking", "stem", "az", "cent", "\u2581ob", "\u2581ass", "land", "\u2581doesn", "\u2581business", "\u2581course", "\u2581ten", "ps", "arch", "ced", "ms", "ize", "nce", "\u2581ref", "\u2581name", "ross", "\u2581grow", "oney", "\u2581went", "ics", "teen", "\u2581cou", "\u2581prob", "\u2581ret", "\u2581guys", "\u2581came", "ash", "led", "\u2581Eur", "ues", "\u2581ide", "gan", "\u2581everything", "\u2581getting", "\u2581ask", "\u2581cor", "\u2581build", "\u2581sign", "\u2581small", "uck", "\u2581el", "\u2581col", "\u2581Is", "ational", "stand", "cy", "\u2581conf", "der", "\u2581bre", "\u2581cap", "\u2581mod", "ets", "ike", "\u2581number", "\u2581comple", "ertain", "\u2581ever", "\u2581coll", "\u2581hum", "\u2581Europe", "\u2581cre", "\u2581met", "\u2581exam", "\u2581move", "\u2581pass", "\u2581left", "\u2581system", "\u2581includ", "\u2581Thank", "cept", "\u2581wom", "\u2581product", "ten", "\u2581rest", "\u2581probably", "\u2581dri", "\u2581Do", "\u2581gener", "\u2581anything", "\u2581lar", "\u2581My", "\u2581school", "\u2581lead", "\u2581sub", "\u2581ty", "\u2581plan", "\u2581seem", "\u2581whole", "irect", "\u2581light", "\u2581must", "\u2581mom", "\u2581opp", "\u2581support", "\u2581family", "ices", "amp", "\u2581proble", "\u2581dr", "ready", "\u2581using", "ense", "\u2581prov", "ush", "ax", "\u2581power", "\u2581Re", "alth", "\u2581ev", "\u2581stand", "\u2581war", "ts", "\u2581", "e", "t", "o", "a", "n", "i", "s", "r", "h", "l", "d", "u", "c", "m", "y", "g", "w", "f", "p", ".", "b", ",", "v", "k", "'", "I", "T", "A", "S", "x", "W", "j", "B", "C", "H", "?", "M", "O", "Y", "N", "P", "E", "q", "L", "D", "z", "G", "F", "R", "!", "J", "U", "K", "V", "Q", "Z", "X" ] }, "decoding": { "strategy": "greedy_batch", "model_type": "tdt", "durations": [ 0, 1, 2, 3, 4 ], "greedy": { "max_symbols": 10 }, "beam": { "beam_size": 2, "return_best_hypothesis": false, "score_norm": true, "tsd_max_sym_exp": 50, "alsd_max_target_len": 2.0 } }, "aux_ctc": { "ctc_loss_weight": 0.3, "use_cer": false, "ctc_reduction": "mean_batch", "decoder": { "_target_": "nemo.collections.asr.modules.ConvASRDecoder", "feat_in": 512, "num_classes": 1024, "vocabulary": [ "", "\u2581t", "\u2581th", "\u2581a", "in", "re", "\u2581the", "\u2581w", "\u2581s", "\u2581o", "er", "ou", "at", "nd", "it", "\u2581h", "\u2581c", "\u2581b", "is", "en", "on", "ing", "\u2581f", "\u2581to", "\u2581m", "es", "\u2581p", "or", "an", "\u2581d", "ll", "\u2581I", "ed", "\u2581and", "\u2581l", "\u2581of", "\u2581in", "\u2581y", "ar", "\u2581g", "\u2581you", "as", "om", "\u2581n", "ve", "\u2581that", "le", "ic", "us", "ow", "et", "al", "\u2581e", "ut", "\u2581it", "ot", "\u2581be", "\u2581T", "ion", "\u2581is", "\u2581wh", "\u2581re", "\u2581on", "\u2581we", "ent", "\u2581A", "ay", "\u2581ha", "\u2581Th", "id", "\u2581S", "ac", "gh", "ver", "ke", "\u2581for", "im", "ly", "ur", "ld", "\u2581he", "\u2581st", "all", "ro", "st", "se", "ct", "ith", "ir", "am", "\u2581this", "if", "\u2581W", "oo", "ri", "\u2581was", "ght", "\u2581u", "\u2581with", "ad", "ch", "\u2581se", "\u2581k", "\u2581an", "\u2581The", "\u2581li", "\u2581do", "\u2581B", "\u2581have", "\u2581as", "th", "\u2581are", "\u2581sh", "ust", "ce", "ally", "ill", "\u2581H", "\u2581j", "ter", "\u2581go", "\u2581And", "ation", "\u2581C", "\u2581so", "ome", "\u2581not", "op", "il", "ore", "\u2581ne", "\u2581can", "\u2581me", "\u2581at", "ould", "ant", "\u2581M", "\u2581like", "ere", "\u2581they", "ra", "ers", "\u2581ab", "\u2581de", "\u2581kn", "ge", "\u2581Y", "\u2581ch", "ul", "pp", "\u2581or", "\u2581al", "\u2581con", "\u2581com", "ess", "\u2581su", "out", "\u2581your", "\u2581So", "ate", "\u2581one", "\u2581all", "\u2581ex", "est", "\u2581fr", "\u2581just", "\u2581pro", "\u2581know", "\u2581O", "ain", "\u2581but", "ol", "ive", "\u2581v", "use", "very", "art", "qu", "\u2581my", "el", "\u2581N", "nt", "\u2581It", "\u2581what", "ab", "\u2581P", "\u2581wor", "\u2581out", "\u2581there", "\u2581up", "um", "\u2581from", "pe", "\u2581tw", "\u2581r", "and", "ight", "ort", "un", "\u2581L", "ist", "\u2581about", "ide", "ig", "ake", "\u2581D", "em", "os", "king", "rou", "ind", "our", "res", "\u2581We", "\u2581get", "\u2581E", "\u2581G", "ack", "\u2581le", "ity", "od", "\u2581F", "ard", "\u2581pl", "\u2581our", "\u2581int", "ment", "\u2581will", "ies", "\u2581by", "ink", "ca", "\u2581if", "red", "her", "ie", "\u2581us", "\u2581some", "\u2581don", "ven", "ood", "ast", "\u2581R", "\u2581his", "\u2581tim", "\u2581tr", "\u2581more", "ich", "ous", "ame", "\u2581going", "\u2581had", "\u2581them", "ook", "\u2581pe", "\u2581Wh", "\u2581You", "\u2581But", "ine", "\u2581here", "\u2581would", "cause", "right", "so", "ost", "ure", "\u2581has", "ect", "\u2581think", "\u2581fe", "ong", "\u2581see", "\u2581when", "\u2581who", "\u2581were", "\u2581really", "\u2581their", "\u2581want", "one", "ople", "\u2581then", "\u2581time", "\u2581sa", "ap", "\u2581te", "\u2581He", "\u2581ye", "ck", "\u2581her", "\u2581thing", "\u2581right", "\u2581which", "itt", "ice", "act", "\u2581people", "ty", "\u2581two", "\u2581J", "\u2581im", "ther", "ci", "ose", "\u2581cl", "\u2581qu", "\u2581man", "\u2581also", "ree", "\u2581en", "ud", "\u2581how", "reat", "ak", "hing", "ag", "\u2581any", "ff", "ace", "per", "\u2581because", "\u2581very", "own", "\u2581ad", "\u2581act", "\u2581been", "\u2581now", "\u2581ag", "\u2581into", "\u2581comp", "ars", "ions", "are", "ite", "iv", "\u2581these", "ays", "ep", "\u2581This", "\u2581she", "ans", "ah", "een", "\u2581over", "ry", "\u2581lo", "age", "\u2581pr", "\u2581sp", "ue", "\u2581co", "ick", "ber", "\u2581did", "ip", "ach", "\u2581back", "\u2581no", "\u2581cont", "\u2581other", "\u2581every", "pt", "\u2581need", "\u2581him", "\u2581U", "\u2581In", "\u2581work", "irst", "\u2581part", "\u2581look", "ittle", "ble", "iz", "\u2581un", "\u2581make", "omet", "nder", "ish", "na", "\u2581little", "\u2581off", "\u2581than", "\u2581got", "ually", "\u2581per", "\u2581good", "\u2581way", "\u2581could", "\u2581ac", "\u2581imp", "able", "\u2581where", "iff", "\u2581That", "\u2581res", "ount", "pl", "ance", "\u2581first", "\u2581ro", "\u2581pre", "ass", "\u2581say", "int", "ated", "ire", "uch", "ase", "\u2581somet", "ound", "\u2581down", "\u2581diff", "sel", "\u2581gu", "\u2581am", "ress", "\u2581lot", "ence", "\u2581dis", "orm", "ix", "\u2581po", "ving", "enty", "\u2581K", "\u2581spe", "und", "he", "\u2581much", "\u2581ar", "round", "\u2581app", "co", "ark", "\u2581new", "ater", "ult", "end", "\u2581even", "\u2581start", "ations", "rough", "ile", "fter", "\u2581well", "be", "\u2581They", "\u2581three", "ign", "ild", "\u2581said", "ough", "ang", "\u2581too", "ade", "\u2581bl", "ens", "\u2581inc", "ia", "\u2581those", "\u2581mo", "\u2581take", "\u2581through", "\u2581fl", "\u2581kind", "\u2581things", "\u2581bet", "\u2581only", "\u2581St", "\u2581let", "cess", "\u2581Ch", "ary", "vel", "\u2581If", "xt", "other", "av", "ical", "ord", "\u2581again", "\u2581something", "onna", "fore", "\u2581may", "ting", "\u2581bu", "\u2581differe", "urn", "\u2581gonna", "\u2581does", "uct", "og", "\u2581twenty", "\u2581gr", "\u2581Ye", "wn", "\u2581should", "\u2581comm", "ition", "\u2581under", "\u2581hel", "ory", "\u2581fo", "\u2581use", "igh", "ife", "\u2581actually", "\u2581tal", "\u2581call", "ents", "ious", "ull", "\u2581There", "\u2581Yeah", "\u2581most", "\u2581ke", "ors", "ved", "ys", "\u2581sc", "\u2581happ", "ope", "\u2581help", "atch", "\u2581What", "\u2581rem", "ple", "\u2581Now", "\u2581br", "ool", "oth", "\u2581four", "self", "\u2581str", "ne", "thing", "\u2581put", "ial", "\u2581great", "ail", "ub", "ning", "\u2581sm", "\u2581feel", "\u2581five", "ody", "undred", "iss", "ank", "get", "aking", "\u2581many", "\u2581hundred", "\u2581years", "\u2581being", "\u2581come", "\u2581mean", "ily", "\u2581different", "\u2581after", "\u2581ser", "\u2581show", "form", "ful", "oy", "\u2581six", "\u2581vide", "\u2581V", "\u2581its", "\u2581point", "\u2581day", "\u2581des", "ons", "\u2581bit", "\u2581bel", "\u2581before", "\u2581aw", "\u2581end", "\u2581Oh", "\u2581still", "ath", "\u2581long", "\u2581'", "ise", "ob", "day", "\u2581add", "ft", "ves", "ces", "ady", "\u2581cr", "\u2581around", "\u2581try", "les", "vers", "kay", "ian", "ates", "\u2581find", "ward", "\u2581As", "\u2581eight", "lic", "\u2581same", "\u2581pos", "\u2581em", "\u2581made", "\u2581supp", "\u2581life", "\u2581Be", "pect", "\u2581dec", "\u2581play", "ange", "\u2581att", "\u2581pers", "ways", "\u2581high", "\u2581hand", "\u2581next", "\u2581cons", "\u2581own", "\u2581inv", "ower", "\u2581ind", "ert", "ng", "ave", "\u2581year", "\u2581big", "ating", "\u2581world", "\u2581rel", "\u2581sure", "\u2581tra", "ew", "ered", "\u2581fin", "\u2581Well", "\u2581sl", "\u2581doing", "bs", "\u2581set", "\u2581rec", "ual", "cial", "\u2581ph", "erm", "\u2581love", "ph", "\u2581real", "\u2581last", "ict", "\u2581bo", "\u2581ra", "ible", "\u2581wr", "mer", "\u2581count", "ities", "\u2581always", "inet", "ments", "uc", "\u2581might", "\u2581inter", "\u2581video", "gin", "\u2581tell", "\u2581never", "vent", "\u2581import", "ied", "\u2581sy", "\u2581How", "ically", "ought", "\u2581thir", "\u2581rep", "ks", "ib", "\u2581fam", "ject", "\u2581bas", "\u2581She", "\u2581give", "akes", "\u2581ninet", "\u2581reg", "\u2581min", "\u2581op", "\u2581def", "\u2581didn", "te", "\u2581cour", "\u2581why", "\u2581ent", "\u2581place", "\u2581ins", "\u2581car", "ather", "\u2581person", "ular", "\u2581inst", "\u2581prod", "lect", "\u2581Al", "\u2581today", "\u2581bec", "\u2581sur", "\u2581All", "\u2581another", "\u2581bus", "\u2581keep", "ell", "ese", "riend", "\u2581quest", "\u2581talk", "als", "ings", "\u2581mon", "cond", "old", "\u2581acc", "\u2581la", "\u2581num", "ident", "\u2581che", "iness", "\u2581turn", "\u2581ear", "\u2581No", "ousand", "\u2581better", "ific", "\u2581loo", "\u2581gl", "oc", "\u2581important", "ited", "\u2581An", "\u2581thousand", "ility", "llow", "\u2581used", "\u2581gen", "\u2581sim", "li", "\u2581happen", "\u2581Un", "\u2581Let", "air", "ock", "ably", "gg", "\u2581watch", "\u2581For", "\u2581sw", "ren", "ute", "ever", "\u2581pol", "\u2581sch", "\u2581When", "\u2581such", "\u2581fif", "\u2581home", "\u2581cle", "\u2581contin", "ouse", "\u2581friend", "uring", "\u2581Okay", "gr", "\u2581able", "\u2581stud", "\u2581eff", "hip", "body", "\u2581top", "ness", "\u2581exper", "\u2581pret", "\u2581both", "\u2581done", "cri", "\u2581mark", "\u2581while", "\u2581old", "ros", "ont", "\u2581second", "ative", "\u2581thought", "\u2581best", "\u2581found", "iew", "\u2581belie", "\u2581each", "erest", "\u2581tri", "\u2581eas", "\u2581ca", "\u2581fact", "\u2581care", "\u2581fun", "atter", "ures", "\u2581head", "\u2581lear", "\u2581water", "\u2581hard", "\u2581few", "\u2581side", "ween", "\u2581exp", "\u2581away", "its", "\u2581ext", "lud", "\u2581run", "\u2581trans", "ince", "\u2581sk", "\u2581open", "cus", "\u2581between", "\u2581called", "\u2581wee", "\u2581pretty", "ason", "\u2581far", "ember", "omm", "\u2581interest", "any", "ner", "uff", "\u2581pres", "\u2581cur", "\u2581child", "ee", "\u2581toget", "\u2581together", "olog", "\u2581God", "ond", "\u2581char", "\u2581looking", "stem", "az", "cent", "\u2581ob", "\u2581ass", "land", "\u2581doesn", "\u2581business", "\u2581course", "\u2581ten", "ps", "arch", "ced", "ms", "ize", "nce", "\u2581ref", "\u2581name", "ross", "\u2581grow", "oney", "\u2581went", "ics", "teen", "\u2581cou", "\u2581prob", "\u2581ret", "\u2581guys", "\u2581came", "ash", "led", "\u2581Eur", "ues", "\u2581ide", "gan", "\u2581everything", "\u2581getting", "\u2581ask", "\u2581cor", "\u2581build", "\u2581sign", "\u2581small", "uck", "\u2581el", "\u2581col", "\u2581Is", "ational", "stand", "cy", "\u2581conf", "der", "\u2581bre", "\u2581cap", "\u2581mod", "ets", "ike", "\u2581number", "\u2581comple", "ertain", "\u2581ever", "\u2581coll", "\u2581hum", "\u2581Europe", "\u2581cre", "\u2581met", "\u2581exam", "\u2581move", "\u2581pass", "\u2581left", "\u2581system", "\u2581includ", "\u2581Thank", "cept", "\u2581wom", "\u2581product", "ten", "\u2581rest", "\u2581probably", "\u2581dri", "\u2581Do", "\u2581gener", "\u2581anything", "\u2581lar", "\u2581My", "\u2581school", "\u2581lead", "\u2581sub", "\u2581ty", "\u2581plan", "\u2581seem", "\u2581whole", "irect", "\u2581light", "\u2581must", "\u2581mom", "\u2581opp", "\u2581support", "\u2581family", "ices", "amp", "\u2581proble", "\u2581dr", "ready", "\u2581using", "ense", "\u2581prov", "ush", "ax", "\u2581power", "\u2581Re", "alth", "\u2581ev", "\u2581stand", "\u2581war", "ts", "\u2581", "e", "t", "o", "a", "n", "i", "s", "r", "h", "l", "d", "u", "c", "m", "y", "g", "w", "f", "p", ".", "b", ",", "v", "k", "'", "I", "T", "A", "S", "x", "W", "j", "B", "C", "H", "?", "M", "O", "Y", "N", "P", "E", "q", "L", "D", "z", "G", "F", "R", "!", "J", "U", "K", "V", "Q", "Z", "X" ] }, "decoding": { "strategy": "greedy", "greedy": {} } }, "interctc": { "loss_weights": [], "apply_at_layers": [] }, "loss": { "loss_name": "tdt", "tdt_kwargs": { "fastemit_lambda": 0.0, "clamp": -1.0, "durations": [ 0, 1, 2, 3, 4 ], "sigma": 0.02, "omega": 0.1 } }, "optim": { "name": "adamw", "lr": 2.0, "betas": [ 0.9, 0.98 ], "weight_decay": 0.001, "sched": { "name": "NoamAnnealing", "d_model": 512, "warmup_steps": 5000, "warmup_ratio": null, "min_lr": 1e-06 } }, "data": { "train_ds": { "shuffle_buffer_size": 20000, "buffer_size": 20000 }, "validation_ds": { "num_workers": 4 } }, "labels": [ "", "\u2581t", "\u2581th", "\u2581a", "in", "re", "\u2581the", "\u2581w", "\u2581s", "\u2581o", "er", "ou", "at", "nd", "it", "\u2581h", "\u2581c", "\u2581b", "is", "en", "on", "ing", "\u2581f", "\u2581to", "\u2581m", "es", "\u2581p", "or", "an", "\u2581d", "ll", "\u2581I", "ed", "\u2581and", "\u2581l", "\u2581of", "\u2581in", "\u2581y", "ar", "\u2581g", "\u2581you", "as", "om", "\u2581n", "ve", "\u2581that", "le", "ic", "us", "ow", "et", "al", "\u2581e", "ut", "\u2581it", "ot", "\u2581be", "\u2581T", "ion", "\u2581is", "\u2581wh", "\u2581re", "\u2581on", "\u2581we", "ent", "\u2581A", "ay", "\u2581ha", "\u2581Th", "id", "\u2581S", "ac", "gh", "ver", "ke", "\u2581for", "im", "ly", "ur", "ld", "\u2581he", "\u2581st", "all", "ro", "st", "se", "ct", "ith", "ir", "am", "\u2581this", "if", "\u2581W", "oo", "ri", "\u2581was", "ght", "\u2581u", "\u2581with", "ad", "ch", "\u2581se", "\u2581k", "\u2581an", "\u2581The", "\u2581li", "\u2581do", "\u2581B", "\u2581have", "\u2581as", "th", "\u2581are", "\u2581sh", "ust", "ce", "ally", "ill", "\u2581H", "\u2581j", "ter", "\u2581go", "\u2581And", "ation", "\u2581C", "\u2581so", "ome", "\u2581not", "op", "il", "ore", "\u2581ne", "\u2581can", "\u2581me", "\u2581at", "ould", "ant", "\u2581M", "\u2581like", "ere", "\u2581they", "ra", "ers", "\u2581ab", "\u2581de", "\u2581kn", "ge", "\u2581Y", "\u2581ch", "ul", "pp", "\u2581or", "\u2581al", "\u2581con", "\u2581com", "ess", "\u2581su", "out", "\u2581your", "\u2581So", "ate", "\u2581one", "\u2581all", "\u2581ex", "est", "\u2581fr", "\u2581just", "\u2581pro", "\u2581know", "\u2581O", "ain", "\u2581but", "ol", "ive", "\u2581v", "use", "very", "art", "qu", "\u2581my", "el", "\u2581N", "nt", "\u2581It", "\u2581what", "ab", "\u2581P", "\u2581wor", "\u2581out", "\u2581there", "\u2581up", "um", "\u2581from", "pe", "\u2581tw", "\u2581r", "and", "ight", "ort", "un", "\u2581L", "ist", "\u2581about", "ide", "ig", "ake", "\u2581D", "em", "os", "king", "rou", "ind", "our", "res", "\u2581We", "\u2581get", "\u2581E", "\u2581G", "ack", "\u2581le", "ity", "od", "\u2581F", "ard", "\u2581pl", "\u2581our", "\u2581int", "ment", "\u2581will", "ies", "\u2581by", "ink", "ca", "\u2581if", "red", "her", "ie", "\u2581us", "\u2581some", "\u2581don", "ven", "ood", "ast", "\u2581R", "\u2581his", "\u2581tim", "\u2581tr", "\u2581more", "ich", "ous", "ame", "\u2581going", "\u2581had", "\u2581them", "ook", "\u2581pe", "\u2581Wh", "\u2581You", "\u2581But", "ine", "\u2581here", "\u2581would", "cause", "right", "so", "ost", "ure", "\u2581has", "ect", "\u2581think", "\u2581fe", "ong", "\u2581see", "\u2581when", "\u2581who", "\u2581were", "\u2581really", "\u2581their", "\u2581want", "one", "ople", "\u2581then", "\u2581time", "\u2581sa", "ap", "\u2581te", "\u2581He", "\u2581ye", "ck", "\u2581her", "\u2581thing", "\u2581right", "\u2581which", "itt", "ice", "act", "\u2581people", "ty", "\u2581two", "\u2581J", "\u2581im", "ther", "ci", "ose", "\u2581cl", "\u2581qu", "\u2581man", "\u2581also", "ree", "\u2581en", "ud", "\u2581how", "reat", "ak", "hing", "ag", "\u2581any", "ff", "ace", "per", "\u2581because", "\u2581very", "own", "\u2581ad", "\u2581act", "\u2581been", "\u2581now", "\u2581ag", "\u2581into", "\u2581comp", "ars", "ions", "are", "ite", "iv", "\u2581these", "ays", "ep", "\u2581This", "\u2581she", "ans", "ah", "een", "\u2581over", "ry", "\u2581lo", "age", "\u2581pr", "\u2581sp", "ue", "\u2581co", "ick", "ber", "\u2581did", "ip", "ach", "\u2581back", "\u2581no", "\u2581cont", "\u2581other", "\u2581every", "pt", "\u2581need", "\u2581him", "\u2581U", "\u2581In", "\u2581work", "irst", "\u2581part", "\u2581look", "ittle", "ble", "iz", "\u2581un", "\u2581make", "omet", "nder", "ish", "na", "\u2581little", "\u2581off", "\u2581than", "\u2581got", "ually", "\u2581per", "\u2581good", "\u2581way", "\u2581could", "\u2581ac", "\u2581imp", "able", "\u2581where", "iff", "\u2581That", "\u2581res", "ount", "pl", "ance", "\u2581first", "\u2581ro", "\u2581pre", "ass", "\u2581say", "int", "ated", "ire", "uch", "ase", "\u2581somet", "ound", "\u2581down", "\u2581diff", "sel", "\u2581gu", "\u2581am", "ress", "\u2581lot", "ence", "\u2581dis", "orm", "ix", "\u2581po", "ving", "enty", "\u2581K", "\u2581spe", "und", "he", "\u2581much", "\u2581ar", "round", "\u2581app", "co", "ark", "\u2581new", "ater", "ult", "end", "\u2581even", "\u2581start", "ations", "rough", "ile", "fter", "\u2581well", "be", "\u2581They", "\u2581three", "ign", "ild", "\u2581said", "ough", "ang", "\u2581too", "ade", "\u2581bl", "ens", "\u2581inc", "ia", "\u2581those", "\u2581mo", "\u2581take", "\u2581through", "\u2581fl", "\u2581kind", "\u2581things", "\u2581bet", "\u2581only", "\u2581St", "\u2581let", "cess", "\u2581Ch", "ary", "vel", "\u2581If", "xt", "other", "av", "ical", "ord", "\u2581again", "\u2581something", "onna", "fore", "\u2581may", "ting", "\u2581bu", "\u2581differe", "urn", "\u2581gonna", "\u2581does", "uct", "og", "\u2581twenty", "\u2581gr", "\u2581Ye", "wn", "\u2581should", "\u2581comm", "ition", "\u2581under", "\u2581hel", "ory", "\u2581fo", "\u2581use", "igh", "ife", "\u2581actually", "\u2581tal", "\u2581call", "ents", "ious", "ull", "\u2581There", "\u2581Yeah", "\u2581most", "\u2581ke", "ors", "ved", "ys", "\u2581sc", "\u2581happ", "ope", "\u2581help", "atch", "\u2581What", "\u2581rem", "ple", "\u2581Now", "\u2581br", "ool", "oth", "\u2581four", "self", "\u2581str", "ne", "thing", "\u2581put", "ial", "\u2581great", "ail", "ub", "ning", "\u2581sm", "\u2581feel", "\u2581five", "ody", "undred", "iss", "ank", "get", "aking", "\u2581many", "\u2581hundred", "\u2581years", "\u2581being", "\u2581come", "\u2581mean", "ily", "\u2581different", "\u2581after", "\u2581ser", "\u2581show", "form", "ful", "oy", "\u2581six", "\u2581vide", "\u2581V", "\u2581its", "\u2581point", "\u2581day", "\u2581des", "ons", "\u2581bit", "\u2581bel", "\u2581before", "\u2581aw", "\u2581end", "\u2581Oh", "\u2581still", "ath", "\u2581long", "\u2581'", "ise", "ob", "day", "\u2581add", "ft", "ves", "ces", "ady", "\u2581cr", "\u2581around", "\u2581try", "les", "vers", "kay", "ian", "ates", "\u2581find", "ward", "\u2581As", "\u2581eight", "lic", "\u2581same", "\u2581pos", "\u2581em", "\u2581made", "\u2581supp", "\u2581life", "\u2581Be", "pect", "\u2581dec", "\u2581play", "ange", "\u2581att", "\u2581pers", "ways", "\u2581high", "\u2581hand", "\u2581next", "\u2581cons", "\u2581own", "\u2581inv", "ower", "\u2581ind", "ert", "ng", "ave", "\u2581year", "\u2581big", "ating", "\u2581world", "\u2581rel", "\u2581sure", "\u2581tra", "ew", "ered", "\u2581fin", "\u2581Well", "\u2581sl", "\u2581doing", "bs", "\u2581set", "\u2581rec", "ual", "cial", "\u2581ph", "erm", "\u2581love", "ph", "\u2581real", "\u2581last", "ict", "\u2581bo", "\u2581ra", "ible", "\u2581wr", "mer", "\u2581count", "ities", "\u2581always", "inet", "ments", "uc", "\u2581might", "\u2581inter", "\u2581video", "gin", "\u2581tell", "\u2581never", "vent", "\u2581import", "ied", "\u2581sy", "\u2581How", "ically", "ought", "\u2581thir", "\u2581rep", "ks", "ib", "\u2581fam", "ject", "\u2581bas", "\u2581She", "\u2581give", "akes", "\u2581ninet", "\u2581reg", "\u2581min", "\u2581op", "\u2581def", "\u2581didn", "te", "\u2581cour", "\u2581why", "\u2581ent", "\u2581place", "\u2581ins", "\u2581car", "ather", "\u2581person", "ular", "\u2581inst", "\u2581prod", "lect", "\u2581Al", "\u2581today", "\u2581bec", "\u2581sur", "\u2581All", "\u2581another", "\u2581bus", "\u2581keep", "ell", "ese", "riend", "\u2581quest", "\u2581talk", "als", "ings", "\u2581mon", "cond", "old", "\u2581acc", "\u2581la", "\u2581num", "ident", "\u2581che", "iness", "\u2581turn", "\u2581ear", "\u2581No", "ousand", "\u2581better", "ific", "\u2581loo", "\u2581gl", "oc", "\u2581important", "ited", "\u2581An", "\u2581thousand", "ility", "llow", "\u2581used", "\u2581gen", "\u2581sim", "li", "\u2581happen", "\u2581Un", "\u2581Let", "air", "ock", "ably", "gg", "\u2581watch", "\u2581For", "\u2581sw", "ren", "ute", "ever", "\u2581pol", "\u2581sch", "\u2581When", "\u2581such", "\u2581fif", "\u2581home", "\u2581cle", "\u2581contin", "ouse", "\u2581friend", "uring", "\u2581Okay", "gr", "\u2581able", "\u2581stud", "\u2581eff", "hip", "body", "\u2581top", "ness", "\u2581exper", "\u2581pret", "\u2581both", "\u2581done", "cri", "\u2581mark", "\u2581while", "\u2581old", "ros", "ont", "\u2581second", "ative", "\u2581thought", "\u2581best", "\u2581found", "iew", "\u2581belie", "\u2581each", "erest", "\u2581tri", "\u2581eas", "\u2581ca", "\u2581fact", "\u2581care", "\u2581fun", "atter", "ures", "\u2581head", "\u2581lear", "\u2581water", "\u2581hard", "\u2581few", "\u2581side", "ween", "\u2581exp", "\u2581away", "its", "\u2581ext", "lud", "\u2581run", "\u2581trans", "ince", "\u2581sk", "\u2581open", "cus", "\u2581between", "\u2581called", "\u2581wee", "\u2581pretty", "ason", "\u2581far", "ember", "omm", "\u2581interest", "any", "ner", "uff", "\u2581pres", "\u2581cur", "\u2581child", "ee", "\u2581toget", "\u2581together", "olog", "\u2581God", "ond", "\u2581char", "\u2581looking", "stem", "az", "cent", "\u2581ob", "\u2581ass", "land", "\u2581doesn", "\u2581business", "\u2581course", "\u2581ten", "ps", "arch", "ced", "ms", "ize", "nce", "\u2581ref", "\u2581name", "ross", "\u2581grow", "oney", "\u2581went", "ics", "teen", "\u2581cou", "\u2581prob", "\u2581ret", "\u2581guys", "\u2581came", "ash", "led", "\u2581Eur", "ues", "\u2581ide", "gan", "\u2581everything", "\u2581getting", "\u2581ask", "\u2581cor", "\u2581build", "\u2581sign", "\u2581small", "uck", "\u2581el", "\u2581col", "\u2581Is", "ational", "stand", "cy", "\u2581conf", "der", "\u2581bre", "\u2581cap", "\u2581mod", "ets", "ike", "\u2581number", "\u2581comple", "ertain", "\u2581ever", "\u2581coll", "\u2581hum", "\u2581Europe", "\u2581cre", "\u2581met", "\u2581exam", "\u2581move", "\u2581pass", "\u2581left", "\u2581system", "\u2581includ", "\u2581Thank", "cept", "\u2581wom", "\u2581product", "ten", "\u2581rest", "\u2581probably", "\u2581dri", "\u2581Do", "\u2581gener", "\u2581anything", "\u2581lar", "\u2581My", "\u2581school", "\u2581lead", "\u2581sub", "\u2581ty", "\u2581plan", "\u2581seem", "\u2581whole", "irect", "\u2581light", "\u2581must", "\u2581mom", "\u2581opp", "\u2581support", "\u2581family", "ices", "amp", "\u2581proble", "\u2581dr", "ready", "\u2581using", "ense", "\u2581prov", "ush", "ax", "\u2581power", "\u2581Re", "alth", "\u2581ev", "\u2581stand", "\u2581war", "ts", "\u2581", "e", "t", "o", "a", "n", "i", "s", "r", "h", "l", "d", "u", "c", "m", "y", "g", "w", "f", "p", ".", "b", ",", "v", "k", "'", "I", "T", "A", "S", "x", "W", "j", "B", "C", "H", "?", "M", "O", "Y", "N", "P", "E", "q", "L", "D", "z", "G", "F", "R", "!", "J", "U", "K", "V", "Q", "Z", "X" ], "target": "nemo.collections.asr.models.hybrid_rnnt_ctc_bpe_models.EncDecHybridRNNTCTCBPEModel", "nemo_version": "1.23.0" }