o i2@sdZddlZddlZddlmZmZmZmZmZm Z ddl Z ddl Z ddl mZmZddlmZddlmZddlmZddlmZdd lmZdd lmZdd lmZdd lmZdd l m!Z!ddl"m#Z#ddl$m%Z%ddl&m'Z'ddl(m)Z)ddl*m+Z+ddl,m-Z-ddl.m/Z/ddl0m1Z1ddl2m3Z3ddl4m5Z5ddl6m7Z7ddl8m9Z9ddl:m;Z;ddlZ>m?Z?m@Z@e1deAeededdZBe1d eAed!edd"d#ZCe1d$eAe)e+d%e'd&d"d#ZDe1d'eAeed(ed)d*ZEGd+d,d,e-ZFdS)-zASR Transducer Task.N)Callable CollectionDictListOptionalTuple)check_argument_typescheck_return_type) AbsFrontend)DefaultFrontend) SlidingWindow) AbsSpecAug)SpecAug) AbsDecoder) RNNDecoder)StatelessDecoder)Encoder)ESPnetASRTransducerModel) JointNetwork) AbsNormalize) GlobalMVN) UtteranceMVN)AbsTask) g2p_choices) ClassChoices)CommonCollateFn)CommonPreprocessor)Trainer)get_default_kwargs)NestedDictAction) float_or_none int_or_nonestr2bool str_or_nonefrontend)defaultsliding_windowr%)nameclasses type_checkr%specaug)r*T)r(r)r%optional normalize) global_mvn utterance_mvnr.decoder)rnn statelessr0)r(r)r%c @s8eZdZUdZdZeed<eee e gZ e Z edejfddZedejded eeeeeeejffgeeeeeejffffd d Zedejded eeeeeej fgeeejfffd d Z!e ddeded eedffddZ"e ddeded eedffddZ#edejd e$fddZ%dS)ASRTransducerTaskzASR Transducer Task definition.num_optimizersparsercCs|jdd}|d}|dg7}|jdtddd|jd tdd d|jd tdd d|jd tttdd|jdtidd|jdtidd|jdd}|jdtddd|jdt dgddd|jdtddd|jdtd d!|jd"tgd#dd$d%|jd&tt dd'd%|jd(t dd)d|jd*tdd+d|jd,t d-d.d|jd/tdd0d|jd1t d-d2d|jd3t d4d5d|j D]}||qdS)6zAdd Transducer task arguments. Args: cls: ASRTransducerTask object. parser: Transducer arguments parser. z Task related.) descriptionrequired token_listz --token_listNz!Integer-string mapper for tokens.)typer%helpz --input_sizez,The number of dimensions for input features.z--initz$Type of model initialization to use.z --model_confz*The keyword arguments for the model class.)actionr%r:z--encoder_confz,The keyword arguments for the encoder class.z--joint_network_confz2The keyword arguments for the joint network class.zPreprocess related.z--use_preprocessorTz-Whether to apply preprocessing to input data.z --token_typebpe)r<charwordphnz.The type of tokens to use during tokenization.)r9r%choicesr:z --bpemodelz$The path of the sentencepiece model.z--non_linguistic_symbolsz'The 'non_linguistic_symbols' file path.)r9r:z --cleaner)Ntacotronjaconv vietnamesezText cleaner to use.)r9r@r%r:z--g2pz&g2p method to use if --token_type=phn.z--speech_volume_normalizez2Normalization value for maximum amplitude scaling.z --rir_scpzThe RIR SCP file path.z--rir_apply_prob?z/The probability of the applied RIR convolution.z --noise_scpzThe path of noise SCP file.z--noise_apply_probz.The probability of the applied noise addition.z--noise_db_range13_15z%The range of the noise decibel level.)add_argument_group get_default add_argumentr#r!rrrr"strrr floatclass_choices_list add_arguments)clsr5groupr7 class_choicesrPP/home/ubuntu/.local/lib/python3.10/site-packages/espnet2/tasks/asr_transducer.pyadd_task_argumentsXs     z$ASRTransducerTask.add_task_argumentsargstrainreturncCstsJtdddS)zBuild collate function. Args: cls: ASRTransducerTask object. args: Task arguments. train: Training mode. Return: : Callable collate function. g)float_pad_value int_pad_value)rr)rMrSrTrPrPrQbuild_collate_fns  z"ASRTransducerTask.build_collate_fncCstsJ|jrPt||j|j|j|j|j|jt |dr|j ndt |dr'|j ndt |dr0|j ndt |dr9|j ndt |drB|jndt |drK|jndd }nd}t|sXJ|S) zBuild pre-processing function. Args: cls: ASRTransducerTask object. args: Task arguments. train: Training mode. Return: : Callable pre-processing function. rir_scpNrir_apply_probrD noise_scpnoise_apply_probnoise_db_rangerE) rT token_typer8bpemodelnon_linguistic_symbols text_cleanerg2p_typerZr[r\r]r^speech_volume_normalize)ruse_preprocessorrr_r8r`racleanerg2phasattrrZr[r\r]r^rdr )rMrSrTretvalrPrPrQbuild_preprocess_fns8  z%ASRTransducerTask.build_preprocess_fnTF inference.cCs|sd}|Sd}|S)zRequired data depending on task mode. Args: cls: ASRTransducerTask object. train: Training mode. inference: Inference mode. Return: retval: Required task data. )speechtext)rlrPrMrTrkrirPrPrQrequired_data_names s z%ASRTransducerTask.required_data_namescCsd}t|sJ|S)zOptional data depending on task mode. Args: cls: ASRTransducerTask object. train: Training mode. inference: Inference mode. Return: retval: Optional task data. rP)r rnrPrPrQoptional_data_names6s z%ASRTransducerTask.optional_data_namesc CstsJt|jtr/t|jdd}dd|D}Wdn1s$wYt||_nt|jttfr=t|j}ntdt|}t d||j duret |j}|d i|j}|}nd}|j }|jdur~t |j}|d i|j} nd} |jdurt |j} | d i|j} nd} t|fi|j} | j} t |j}||fi|j}|j}t|| |fi|j}td |||| | | ||d|j}|j durt!d d t"|sJ|S) zRequired data depending on task mode. Args: cls: ASRTransducerTask object. args: Task arguments. Return: model: ASR Transducer model. zutf-8)encodingcSsg|]}|qSrP)rstrip).0linerPrPrQ Zsz1ASRTransducerTask.build_model..Nztoken_list must be str or listzVocabulary size: ) vocab_sizer8r$r*r,encoderr/ joint_networkzCurrently not supported.z7Initialization part will be reworked in a short future.rP)#r isinstancer8rIopenlisttuple RuntimeErrorlenlogginginfo input_sizefrontend_choices get_classr$ frontend_conf output_sizer*specaug_choices specaug_confr,normalize_choicesnormalize_confr encoder_confdecoder_choicesr/ decoder_confrjoint_network_confr model_confinitNotImplementedErrorr )rMrSfr8rvfrontend_classr$r specaug_classr*normalize_classr,rwencoder_output_size decoder_classr/decoder_output_sizerxmodelrPrPrQ build_modelJsx              zASRTransducerTask.build_modelN)TF)&__name__ __module__ __qualname____doc__r4int__annotations__rrrrrKrtrainer classmethodargparseArgumentParserrR NamespaceboolrrrrIrnpndarrayrtorchTensorrYrarrayrjrorprrrPrPrPrQr2Jsh   & .    r2)Grrrtypingrrrrrrnumpyrr typeguardrr !espnet2.asr.frontend.abs_frontendr espnet2.asr.frontend.defaultr espnet2.asr.frontend.windowingr espnet2.asr.specaug.abs_specaugr espnet2.asr.specaug.specaugr*espnet2.asr_transducer.decoder.abs_decoderr*espnet2.asr_transducer.decoder.rnn_decoderr0espnet2.asr_transducer.decoder.stateless_decoderr&espnet2.asr_transducer.encoder.encoderr.espnet2.asr_transducer.espnet_transducer_modelr$espnet2.asr_transducer.joint_networkrespnet2.layers.abs_normalizerespnet2.layers.global_mvnrespnet2.layers.utterance_mvnrespnet2.tasks.abs_taskrespnet2.text.phoneme_tokenizerrespnet2.train.class_choicesrespnet2.train.collate_fnrespnet2.train.preprocessorrespnet2.train.trainerr espnet2.utils.get_default_kwargsr espnet2.utils.nested_dict_actionrespnet2.utils.typesr r!r"r#dictrrrrr2rPrPrPrQs