o i)@sdZddlZddlZddlZddlZddlZddlZddlZddl m Z ddl m Z ddl mZddlmZddZd d Zed krMeejd ddSdS) z%Text-to-speech model training script.N) __version__) TTSInterface) strtobool)BATCH_COUNT_CHOICEScCstjdtjtjd}|jdddd|jdddd|jd dd d|jd d td d|jddtddgdd|jdtddd|jddtdd|jddtdd|jdddtdd d!|jd"d#td$d%d&|jd'd(d)td*d|jd+d tdd,d!|jd-dtd.d|jd/dtd0d|jd1d2td3d|jd4tdd5d|jd6tdd7d|jd8td9d:d&|jd;d)tddtgd?dd@dA|jdBdCtdDdE|jdFdGdHd)tdId|jdJd)tdKd|jdLd)tdMd|jdNd)tdOd|jdPd)tdQd|jdRdSd2tdTdUdV|jdWdXdYtdTdZdV|jd[d)td\d|jd]td d^d&|jd_d`t dad|jdbd`t dcd|jdddetdedfgdgd|jdhdtdid|jdjdkt dld|jdmdnt dod|jdpdnt dqd|jdrdsdttdud|jdvdwtddxd!|jdydztdd{d!|jd|dt d}d|jd~dtdd|jdd`t dd|jdd tdd|jdddddd|jdd tdd|jdddddd|jdd dddd|S)z!Get parser of training arguments.zGTrain a new text-to-speech (TTS) model on one CPU, one or multiple GPUs) descriptionconfig_file_parser_classformatter_classz--configTzconfig file path)is_config_filehelpz --config2zCsecond config file path that overwrites the settings in `--config`.z --config3zRthird config file path that overwrites the settings in `--config` and `--config2`.z--ngpuNz5Number of GPUs. If not given, use all visible devices)defaulttyper z --backendpytorchchainerzBackend library)r r choicesr z--outdirzOutput directory)r requiredr z --debugmode Debugmodez--seedz Random seedz--resumez-r?z!Resume the training from snapshot)r r nargsr z --minibatchesz-Nz-1z&Process only N minibatches (for debug))r r r z --verbosez-VrzVerbose optionz--tensorboard-dirzTensorboard log directory pathz--eval-interval-epochszEvaluation interval epochsz--save-interval-epochszSave interval epochsz--report-interval-itersdzReport interval iterationsz --train-jsonzFilename of training jsonz --valid-jsonzFilename of validation jsonz--model-modulez7espnet.nets.pytorch_backend.e2e_tts_tacotron2:Tacotron2zmodel defined modulez --sortagradzFHow many epochs to use sortagrad for. 0 = deactivated, -1 = all epochsz--batch-sort-keyshuffle)routputinputz@Batch sorting key. "shuffle" only work with --batch-count "seq".)r r rrr z --batch-countautozKHow to count batch_size. The default (auto) will find how to count by args.)r rr z --batch-sizez --batch-seqsz-bz*Maximum seqs in a minibatch (0 to disable)z --batch-binsz*Maximum bins in a minibatch (0 to disable)z--batch-frames-inz2Maximum input frames in a minibatch (0 to disable)z--batch-frames-outz3Maximum output frames in a minibatch (0 to disable)z--batch-frames-inoutz9Maximum input+output frames in a minibatch (0 to disable)z --maxlen-inz--batch-seq-maxlen-inMLzPWhen --batch-count=seq, batch size is reduced if the input sequence length > ML.)r r metavarr z --maxlen-outz--batch-seq-maxlen-outzPWhen --batch-count=seq, batch size is reduced if the output sequence length > MLz--num-iter-processeszNumber of processes of iteratorz--preprocess-confz-The configuration file for the pre-processingz--use-speaker-embeddingFz Whether to use speaker embeddingz--use-second-targetzWhether to use second targetz--optadamnoam Optimizerz --accum-gradzNumber of gradient accumurationz--lrgMbP?zLearning rate for optimizerz--epsgư>zEpsilon for optimizerz--weight-decayz&Weight decay coefficient for optimizerz--epochsz-ezNumber of maximum epochsz--early-stop-criterionzvalidation/main/lossz=Value to monitor to trigger an early stopping of the trainingz --patiencezINumber of epochs to wait without improvement before stopping the trainingz --grad-clipzGradient norm threshold to clipz--num-save-attentionz*Number of samples of attention to be savedz--keep-all-data-on-memz"Whether to keep all data on memoryz --enc-initz1Pre-trained TTS model path to initialize encoder.z--enc-init-modszenc.cfdddDS)Ncg|] }dkrt|qSrstr.0modsH/home/ubuntu/.local/lib/python3.10/site-packages/espnet/bin/tts_train.py 0get_parser....,splitr,r.r,r/ zget_parser..z%(asctime)s (%(module)s:%(lineno)d) %(levelname)s: %(message)s)levelformatzSkip DEBUG/INFO messagesNCUDA_VISIBLE_DEVICESr3z CUDA_VISIBLE_DEVICES is not set.z nvidia-smiz-L)stdoutstderr rzngpu: zrandom seed = %dr )trainzOnly pytorch is supported.)&rAparse_known_argsespnet.utils.dynamic_importrB model_module issubclassr add_arguments parse_argsrversionverboselogging basicConfigINFOWARNwarningngpuosenvirongetlenr5 subprocessrunPIPECalledProcessErrorFileNotFoundErrorrGdecodeinfoseedrandomnpbackendespnet.tts.pytorch_backend.ttsrINotImplementedError) cmd_argsr@args_rB model_classcvdrWprIr.r.r/main&sR              ro__main__r)__doc__rRrXrdr\sysr8numpyreespnetrespnet.nets.tts_interfacerespnet.utils.cli_utilsrespnet.utils.training.batchfyrrAro__name__argvr.r.r.r/s&    @