o j7 @s:ddlZddlZddlZddlZddlZddlZddlZ ddlZddl Z ddl Z ddl Z ddl mZddlmZddlmZddlmZddZdadd Zd d ZGd d d Zee_eje_ddlZejejdeejgdddZ e!dkrej"dej#dZ$e$j%de&ddde$j%de'ddde$j%de&ddde$j%de&d d!de$j%d"e&d#d$de$j%d%e&d&d'de$(Z)e*Z+e,d(Gej-d)e)j.d*Z.ej/d+d,d-Z0ej1d.d/gd0d1Z2ej3d2d3Z4ej-d4d5d6Z5e+j6e de5d$d7ej7d8d9Z8ej9d:d;fd`ej-d?d3Z;ej-d@d3ZdCZ>ej=dDdEdFd$e)j?dCZ?ej=dGdEdHd$e)j@dCZ@ej=dIdEdJd$e)jAdCZAej3d2d3ZBej-d4d5d6ZCe+j6e deCd$d7ej7dKd9ZDdLdMZEWdn 1swYe,dNeFeG)ZHej-dOdPd*ZIej-dQdPd*ZJej-dRdPd*ZKej3d2d3ZLej7dSd9ZMWdn 1swYeG&ZNej-dTdPd*ZOej1dUd/gd0d1ZPej-dVdWd*ZQej7dXd9ZRWdn 1swYeGZSej3d2d3ZTejUdYd3ZVejUdZd3ZWWdn 1swYWdn 1s-wYe8jXe:e0e2e.ge4e;ee?e@e.eAgeBeJeKeIeOgd[eMjXeeIeJeKgeLgd[eRjXeePeQeOgeTeVeWgd[Wdn 1szwYWdn 1swYe+jYd:d5e)jZd\d]dSdS)^N)format_audio_list) train_gpt) XttsConfig)XttscCstjr tjdSdSN)torchcuda is_available empty_cacher r S/home/kuhnn/.local/lib/python3.10/site-packages/TTS/demos/xtts_ft_demo/xtts_demo.pyclear_gpu_caches r cCsht|r |r |s dSt}||t|atdtj|||ddtj r.t tddS)NzYou need to run the previous steps or manually set the `XTTS checkpoint path`, `XTTS config path`, and `XTTS vocab path` fields !!zLoading XTTS model! F)checkpoint_path vocab_path use_deepspeedz Model Loaded!) r r load_jsonrinit_from_config XTTS_MODELprintload_checkpointrrr )xtts_checkpoint xtts_config xtts_vocabconfigr r r load_models    rc Cstdus|sdStj|tjjtjjtjjd\}}tj||||tjjtjjtjj tjj tjj d }t j ddd }t|dd|d<|j}t||dd Wdn1s[wYd ||fS) N)z6You need to run the previous step to load the model !!NN) audio_path gpt_cond_lenmax_ref_lengthsound_norm_refs) textlanguagegpt_cond_latentspeaker_embedding temperaturelength_penaltyrepetition_penaltytop_ktop_pz.wavF)suffixdeletewavri]zSpeech generated !)rget_conditioning_latentsrr max_ref_lenr inferencer#r$r%r&r'tempfileNamedTemporaryFilertensor unsqueezename torchaudiosave)langtts_textspeaker_audio_filer!r"outfpout_pathr r r run_tts+s( "  r;c@s.eZdZd ddZddZddZdd Zd S) Loggerlog.outcCs ||_tj|_t|jd|_dS)Nw)log_filesysstdoutterminalopenlog)selffilenamer r r __init__HszLogger.__init__cCs|j||j|dSr)rBwriterD)rEmessager r r rHMs z Logger.writecCs|j|jdSr)rBflushrDrEr r r rJQs z Logger.flushcCsdS)NFr rKr r r isattyUsz Logger.isattyN)r=)__name__ __module__ __qualname__rGrHrJrLr r r r r<Gs   r<z'%(asctime)s [%(levelname)s] %(message)s)levelformathandlerscCsDtjttjjd }|WdS1swYdS)Nr)r@rArJrCr?read)fr r r read_logsgs $rV__main__zrXTTS fine-tuning demo Example runs: python3 TTS/demos/xtts_ft_demo/xtts_demo.py --port ) descriptionformatter_classz--portz*Port to run the gradio demo. Default: 5003i)typehelpdefaultz --out_pathzMOutput path (where data and checkpoints will be saved) Default: /tmp/xtts_ft/z /tmp/xtts_ft/z --num_epochsz&Number of epochs to train. Default: 10 z --batch_sizezBatch size. Default: 4z --grad_acummz#Grad accumulation steps. Default: 1z--max_audio_lengthz0Max permitted audio size in seconds. Default: 11 z1 - Data processingz7Output path (where data and checkpoints will be saved):)labelvaluemultiplezlSelect here the audio files that you want to use for XTTS trainining (Supported formats: wav, mp3, and flac)) file_countrazDataset Languageen)reesfrdeitptpltrrunlcsarzhhukoja)rarbchoicesz Progress:)razLogs:F)ra interactive)everyzStep 1 - Create dataset)rbT) track_tqdmc Csttj|d}tj|dd|durdSz t||||d\}}}Wntt}d|ddfYSt|dkrLd }t ||ddfSt d d ||fS) NdatasetT)exist_ok)zyYou should provide one or multiple audio files! If you provided it, probably the upload of the files is not finished yet!r{)target_languager:gradio_progresszThe data processing was interrupted due an error !! Please check the console to verify the full error message! Error summary: r{xzUThe sum of the duration of the audios that you provided should be at least 2 minutes!zDataset Processed!) r ospathjoinmakedirsr traceback print_exc format_excr) rr r:progress train_meta eval_metaaudio_total_sizeerrorrIr r r preprocess_datasets$  rz2 - Fine-tuning XTTS Encoderz Train CSV:z Eval CSV:zNumber of epochs:d)raminimummaximumsteprbz Batch size:izGrad accumulation steps:z$Max permitted audio size in seconds:zStep 2 - Run the trainingc Cst|r|s dSzt|d}t||||||||d\}} } } } Wntt} d| ddddfYStd|d| td| d| tj | d}t d td || || fS) N)z\You need to run the data processing step or manually set `Train CSV` and `Eval CSV` fields !r{r{r{r{i"V) output_pathmax_audio_lengthzxThe training was interrupted due an error !! Please check the console to check the full error message! Error summary: r{zcp  zbest_model.pthzModel training done!) r intrrrrrsystemrrr)r train_csveval_csv num_epochs batch_size grad_acummrr config_pathoriginal_xtts_checkpoint vocab_fileexp_path speaker_wavrft_xtts_checkpointr r r train_models  &rz 3 - InferencezXTTS checkpoint path:r{zXTTS config path:zXTTS vocab path:z#Step 3 - Load Fine-tuned XTTS modelzSpeaker reference audio:Languagez Input Text.zBThis model sounds really good and above all, it's reasonably fast.zStep 4 - InferencezGenerated Audio.zReference audio used.)fninputsoutputsz0.0.0.0)sharedebug server_port server_name)[argparserr@r.gradiogrlibrosa.displaylibrosanumpynprr3r&TTS.demos.xtts_ft_demo.utils.formatterr&TTS.demos.xtts_ft_demo.utils.gpt_trainrTTS.tts.configs.xtts_configrTTS.tts.models.xttsrr rrr;r<rAstderrlogging basicConfigINFO StreamHandlerrVrMArgumentParserRawTextHelpFormatterparser add_argumentrstr parse_argsargsBlocksdemoTabTextboxr:File upload_fileDropdownr5Label progress_datalogsloadButtonprompt_compute_btnProgressrrrSliderrrrrprogress_trainlogs_tts_train train_btnrRowColumncol1rrr progress_loadload_btncol2speaker_reference_audio tts_languager6tts_btncol3 progress_genAudiotts_output_audioreference_audioclicklaunchportr r r r s           H   C    ! <  |