o jM@sddlZddlZddlmZddlmZddlmZddlZddlm m Z ddl Z ddl mZddlmZddlmZddlmZmZmZdd lmZdd lmZdd lmZdd lmZmZmZdd l m!Z!ddl"m#Z#ddl$m%Z%ddl&m'Z'm(Z(ddl)m*Z*ddl+m,Z,ddZ-d0ddZ.     d1ddZ/d2d d!Z0d3d"d#Z1  d4d$d%Z2d&d'Z3d(d)Z4eGd*d+d+eZ5eGd,d-d-eZ6Gd.d/d/e,Z7dS)5N)contextmanager) dataclass)time)Coqpit)tqdm)TorchMelSpectrogram)denormalize_tacotron_mel load_voicewav_to_univnet_mel) UnifiedVoice)"AudioMiniEncoderWithClassifierHead)CLVP)SpacedDiffusionget_named_beta_schedulespace_timesteps) DiffusionTts)RandomLatentConverter)VoiceBpeTokenizer)VocConfVocType)Wav2VecAlignment)BaseTTScCsR|dd|f}|jd|kr|}|S|jd|kr't|d||jdf}|S)z Utility function for forcing to have the specified sequence length, whether by clipping it or padding it with 0s. .Nr)shapeFpad)tlengthtprJ/home/kuhnn/.local/lib/python3.10/site-packages/TTS/tts/models/tortoise.pypad_or_truncatesr!cCs.|dur ttn|}t|t||S)z Sets the random seeds that tortoise uses to the current time() and returns that seed so results can be reproduced. N)intrtorch manual_seedrandomseedr&rrr deterministic_state's  r(Tddimc Cs&tt||gdddtd||||dS)z_ Helper function to load a GaussianDiffusion instance configured for use as a vocoder. epsilon learned_rangemselinear) use_timestepsmodel_mean_typemodel_var_type loss_typebetasconditioning_freeconditioning_free_ksampler)rrr)trained_diffusion_stepsdesired_diffusion_steps cond_free cond_free_kr8rrr load_discrete_vocoder_diffuser5s r=cudacKs|jd|}|dkrtj|dt|fd}n|dkr-td|}|dd|||f}tdi||dd}|d |S)zm Converts the given conditioning signal to a MEL spectrogram and clips it as expected by the models. rr)rNr) rrrabsr%randintr unsqueezesqueezeto)clip cond_lengthdevicekwargsgap rand_startmel_cliprrr format_conditioningKs rLcCsv||k}t|dkr|rtd|Sd||<|}d||d<|d|jdkr9d|d<d|d<d |d <|S) a] This function performs some padding on coded audio that fixes a mismatch issue between what the diffusion model was trained on and what the autoregressive code generator creates (which has no padding or end). This is highly specific to the DVAE being used, so this particular coding will not necessarily work if used with a different DVAE. This can be inferred by feeding a audio clip padded with lots of zeros on the end through the DVAE and copying out the last few codes. Failing to do this padding will produce speech with a harsh end that sounds like "BLAH" or similar. rzNo stop tokens found in one of the generated voice clips. This typically means the spoken audio is too long. In some cases, the output will still be good, though. Listen to it and if it is missing words, try breaking up your input text.SN-r)nonzerolenprintminitemr)codes stop_tokencomplainstop_token_indicesstmrrr fix_autoregressive_outputYs   r]c CstH|jdddd}|jdd|f}||||d}tj||jd|} |j||| d |i|d } t| d d d d d |fWd S1sOwYd S) zZ Uses the specified diffusion model to convert discrete codes into a spectrogram. r+]"VrdFrGprecomputed_aligned_embeddings)noise model_kwargsprogressN)r#no_gradrtimestep_independentrandnrG sample_loopr) diffusion_modeldiffuserlatentsconditioning_latents temperatureverboseoutput_seq_len output_shapeprecomputed_embeddingsrdmelrrr do_spectrogram_diffusionws" $rucCsptddddddddddddd }|tjtj|d td d | d}t j ||d d}|ddS)a# Returns whether or not Tortoises' classifier thinks the given clip came from Tortoise. :param clip: torch tensor containing audio waveform data (get it from load_audio) :return: True if the clip was classified as coming from Tortoise and false if it was classified as real. r+ir^ rF) spec_dim embedding_dimdepthdownsample_factor resnet_blocks attn_blocksnum_attn_heads base_channelsdropout kernel_sizedistribute_zero_labelzclassifier.pthcpu map_locationrdim) r load_state_dictr#loadospathjoinrGrrBrsoftmax)rE model_dir classifierresultsrrr classify_audio_clips$$ rcCsTtjr(tj\}}|d}d}|dkrd}|S|dkr"d}|S|dkr(d}|S) z Tries to pick a batch size that will fit in your GPU. These sizes aren't guaranteed to work, but they should give you a good shot. i@r+ r^)r#r? is_available mem_get_info)_ available availableGb batch_sizerrr pick_best_batch_size_for_gpus rc@s2eZdZUdZeed<dZeed<dZeed<dS)TortoiseAudioConfigr` sample_rater_diffusion_sample_rateoutput_sample_rateN)__name__ __module__ __qualname__rr"__annotations__rrrrrr rs  rc@seZdZUdZdZeed<dZeed<dZ eed<dZ eed<d Z e ed <d Z e ed <d Ze ed <d Zeed<ejZeed<dZeed<dZeed<dZeed<dZeed<dZeed<dZeed<d Zeed<d Zeed<dZeed<dZeed<dZeed <d!Zeed"<d#Z eed$<d%Z!eed&<dZ"eed'<d(Z#eed)<d*Z$eed+<dZ%eed,<dZ&eed-<d*Z'eed.<d*Z(eed/<d0Z)eed1<d0Z*eed2<d0Z+eed3<d4Z,eed5<d6Z-eed7<d8Z.eed9<d:Z/eed;<d<d:Z2eed?<d@Z3eedA<dZ4eedB<dCZ5eedD<d S)E TortoiseArgsaA dataclass to represent Tortoise model arguments that define the model structure. Args: autoregressive_batch_size (int): The size of the auto-regressive batch. enable_redaction (bool, optional): Whether to enable redaction. Defaults to True. high_vram (bool, optional): Whether to use high VRAM. Defaults to False. kv_cache (bool, optional): Whether to use the kv_cache. Defaults to True. ar_checkpoint (str, optional): The checkpoint for the autoregressive model. Defaults to None. clvp_checkpoint (str, optional): The checkpoint for the ConditionalLatentVariablePerseq model. Defaults to None. diff_checkpoint (str, optional): The checkpoint for the DiffTTS model. Defaults to None. num_chars (int, optional): The maximum number of characters to generate. Defaults to 255. vocoder (VocType, optional): The vocoder to use for synthesis. Defaults to VocConf.Univnet. For UnifiedVoice model: ar_max_mel_tokens (int, optional): The maximum mel tokens for the autoregressive model. Defaults to 604. ar_max_text_tokens (int, optional): The maximum text tokens for the autoregressive model. Defaults to 402. ar_max_conditioning_inputs (int, optional): The maximum conditioning inputs for the autoregressive model. Defaults to 2. ar_layers (int, optional): The number of layers for the autoregressive model. Defaults to 30. ar_model_dim (int, optional): The model dimension for the autoregressive model. Defaults to 1024. ar_heads (int, optional): The number of heads for the autoregressive model. Defaults to 16. ar_number_text_tokens (int, optional): The number of text tokens for the autoregressive model. Defaults to 255. ar_start_text_token (int, optional): The start text token for the autoregressive model. Defaults to 255. ar_checkpointing (bool, optional): Whether to use checkpointing for the autoregressive model. Defaults to False. ar_train_solo_embeddings (bool, optional): Whether to train embeddings for the autoregressive model. Defaults to False. For DiffTTS model: diff_model_channels (int, optional): The number of channels for the DiffTTS model. Defaults to 1024. diff_num_layers (int, optional): The number of layers for the DiffTTS model. Defaults to 10. diff_in_channels (int, optional): The input channels for the DiffTTS model. Defaults to 100. diff_out_channels (int, optional): The output channels for the DiffTTS model. Defaults to 200. diff_in_latent_channels (int, optional): The input latent channels for the DiffTTS model. Defaults to 1024. diff_in_tokens (int, optional): The input tokens for the DiffTTS model. Defaults to 8193. diff_dropout (int, optional): The dropout percentage for the DiffTTS model. Defaults to 0. diff_use_fp16 (bool, optional): Whether to use fp16 for the DiffTTS model. Defaults to False. diff_num_heads (int, optional): The number of heads for the DiffTTS model. Defaults to 16. diff_layer_drop (int, optional): The layer dropout percentage for the DiffTTS model. Defaults to 0. diff_unconditioned_percentage (int, optional): The percentage of unconditioned inputs for the DiffTTS model. Defaults to 0. For ConditionalLatentVariablePerseq model: clvp_dim_text (int): The dimension of the text input for the CLVP module. Defaults to 768. clvp_dim_speech (int): The dimension of the speech input for the CLVP module. Defaults to 768. clvp_dim_latent (int): The dimension of the latent representation for the CLVP module. Defaults to 768. clvp_num_text_tokens (int): The number of text tokens used by the CLVP module. Defaults to 256. clvp_text_enc_depth (int): The depth of the text encoder in the CLVP module. Defaults to 20. clvp_text_seq_len (int): The maximum sequence length of the text input for the CLVP module. Defaults to 350. clvp_text_heads (int): The number of attention heads used by the text encoder in the CLVP module. Defaults to 12. clvp_num_speech_tokens (int): The number of speech tokens used by the CLVP module. Defaults to 8192. clvp_speech_enc_depth (int): The depth of the speech encoder in the CLVP module. Defaults to 20. clvp_speech_heads (int): The number of attention heads used by the speech encoder in the CLVP module. Defaults to 12. clvp_speech_seq_len (int): The maximum sequence length of the speech input for the CLVP module. Defaults to 430. clvp_use_xformers (bool): A flag indicating whether the model uses transformers in the CLVP module. Defaults to True. duration_const (int): A constant value used in the model. Defaults to 102400. r+autoregressive_batch_sizeFenable_redaction high_vramTkv_cacheN ar_checkpointclvp_checkpointdiff_checkpoint num_charsvocoderi\ar_max_mel_tokensiar_max_text_tokensrvar_max_conditioning_inputs ar_layers ar_model_dimrar_headsar_number_text_tokensar_start_text_tokenar_checkpointingar_train_solo_embeddingsdiff_model_channelsrdiff_num_layersradiff_in_channelsr*diff_out_channelsdiff_in_latent_channelsi diff_in_tokensr diff_dropout diff_use_fp16diff_num_headsdiff_layer_dropdiff_unconditioned_percentagei clvp_dim_textclvp_dim_speechclvp_dim_latentclvp_num_text_tokensclvp_text_enc_depthi^clvp_text_seq_len clvp_text_headsi clvp_num_speech_tokensclvp_speech_enc_depthclvp_speech_headsiclvp_speech_seq_lenclvp_use_xformersiduration_const)6rrr__doc__rr"rrboolrrrstrrrrrUnivnetrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrr rsZ  6                                        rcseZdZdZdeffdd ZeddZ   d*d d Zd d Z d+ddZ ddZ                    d,ddZ dd Z d!d"Zed-d$d%Z      d.d&d'Zd(d)ZZS)/TortoiseaTortoise model class. Currently only supports inference. Examples: >>> from TTS.tts.configs.tortoise_config import TortoiseConfig >>> from TTS.tts.models.tortoise import Tortoise >>> config = TortoiseConfig() >>> model = Tortoise.inif_from_config(config) >>> model.load_checkpoint(config, checkpoint_dir="paths/to/models_dir/", eval=True) configcstj|dddd|_||_|jj|_|jj|_|j|_|jj dur&t n|jj |_ |jj |_ t t jr9dnd|_ |j rDt|_t|_t|jj|jj|jj|jj|jj|jj|jj|jj|jj|jjd |_ t!|jj"|jj#|jj$|jj%|jj&|jj'|jj(|jj)|jj*|jj+|jj,d |_-t.|jj/|jj0|jj1|jj2|jj3|jj4|jj5|jj6|jj7|jj8|jj9|jj:d |_;|jj|_|dvsJdtfdd|D}g}|D]}|t|djjdqtj|dd}j }| |}Wd n1sHwYg} j j } |D]}|rct j |dd d n|d} |dkrt| | } t| jd jd } | | qUddlm} |dkrg}t| | jd| D]6}| d d || |d| f}t|| }t|jd jd } |dkr| | q|dkr|| q|dkr| t|dqUtj| dd} j }| | }Wd n1swYWd n 1s wY|r|||| fS||fS)aU Transforms one or more voice_samples into a tuple (autoregressive_conditioning_latent, diffusion_conditioning_latent). These are expressive learned latents that encode aspects of the provided clips like voice, intonation, and acoustic properties. :param voice_samples: List of arbitrary reference clips, which should be *pairs* of torch tensors containing arbitrary kHz waveform data. :param latent_averaging_mode: 0/1/2 for following modes: 0 - latents will be generated as in original tortoise, using ~4.27s from each voice sample, averaging latent across all samples 1 - latents will be generated using (almost) entire voice samples, averaged across all the ~4.27s chunks 2 - latents will be generated using (almost) entire voice samples, averaged per voice sample )rr+rvz0latent_averaging mode has to be one of (0, 1, 2)csg|] }fdd|DqS)csg|]}|jqSr)rDrG).0vrrr sz@Tortoise.get_conditioning_latents...r)rlsrrr rsz5Tortoise.get_conditioning_latents..r)rG mel_norm_filer+rNr`r_F)do_normalizationrG)ceilrv)r#rgappendrLrGrstackrrget_conditioningrr torchaudio functionalresampler!r rDmathrrangermeanr)r voice_samples return_melslatent_averaging_modeoriginal_tortoise auto_condsrar auto_latentdiffusion_conds DURS_CONSTsamplecond_melrtemp_diffusion_condschunkcurrent_samplerdiffusion_latentrrr get_conditioning_latentssh              0 z!Tortoise.get_conditioning_latentscCs|jdur;td|_|jtjtj|j dt ddtd|_ |j tjtj|j dt ddt |t dg| t dgfWdS1s\wYdS)Nrz rlg_auto.pthrrizrlg_diffuser.pthg)r revalrr#rrrrrrGr rgtensorrrrr get_random_conditioning_latentss$  "$z(Tortoise.get_random_conditioning_latentsr%Nc Ks||durdn|}|dur|g}t||\}}nt|\}}|j||f||d|}|d|d|d|d|dd } | S) aSynthesize speech with the given input text. Args: text (str): Input text. config (TortoiseConfig): Config with inference parameters. speaker_id (str): One of the available speaker names. If `random`, it generates a random speaker. voice_dirs (List[str]): List of paths that host reference audio files for speakers. Defaults to None. **kwargs: Inference settings. See `inference()`. Returns: A dictionary of the output values with `wav` as output waveform, `deterministic_seed` as seed used at inference, `text_input` as text token IDs after tokenizer, `voice_samples` as samples used for cloning, `conditioning_latents` as latents used at inference. Nr%)r&rnwavdeterministic_seedtextr&rn)r9r: text_inputsr&rn)r inference_with_config) rr;r speaker_id voice_dirsrHr&rnoutputs return_dictrrr synthesizes& zTortoise.synthesizec Ks|j|j|j|j|j|j|jd}dddddddddddd d dd dd d dddddd dddddd}d|vrK|||d|d|||j |fi|S)zH inference with config #TODO describe in detail )rolength_penaltyrepetition_penaltytop_pr<diffusion_temperaturer8rrr,)num_autoregressive_samplesdiffusion_iterationsr8rrF)rGrHr;rxzdpm++2mrw2`P)rGrHr*r) single_sample ultra_fastultra_fast_old very_fastfastfast_oldstandard high_qualitypreset) rorCrDrEr<rFr8updatepop inference)rr;rrHsettingspresetsrrr r=sR $  zTortoise.inference_with_configr+Tr皙?@rarv?r,c7 st|d}tjdj}t |d}|j ddks&Jd|dur8j |d||d \}}}}n|durA|\}}n \}}|j}|j}t ||||d }j}| jrojd _| jsctg}| j}jj} d }!jj_|rtd jV}"tjdtj|d=tt|| dD]+}#|"j||fd| | j| | |d|}$||$j d}%tj |$d|%f| d}$||$qWdn1swYWdn1swY|_g}&jl}'tjdtj|dRt|| dD]*}(t|(j dD] })t|(|)| |(|)<q|'||(j dd|(dd}*|&|*qtj|&dd}&tj|dd}|tj|&|dj}+Wdn 1s_wYWdn 1sowY~j3}"|"||d||dtj |j dg|jd|+tj |+j djj!g|jdddd},Wdn 1swY~|rtdg}-t|+j dD]}#|+|#d}$|,|#d}.d}/t|$j dD]$}0|$d|0f|!kr|/d7}/nd}/|/dkr|.ddd|0f}.nqj"}1t#|1||.|||d}2Wdn 1s"wYj$}3|3%|2}4|-|4&Wdn 1sEwYqȇfddfdd |-D}-t'|-dkrf|-}5n|-d}5Wdn 1suwY|5ddddd!}6|r|5|||d!}6|6S)"a This function produces an audio clip of the given text being spoken with the given reference voice. Args: text: (str) Text to be spoken. voice_samples: (List[Tuple[torch.Tensor]]) List of an arbitrary number of reference clips, which should be tuple-pairs of torch tensors containing arbitrary kHz waveform data. conditioning_latents: (Tuple[autoregressive_conditioning_latent, diffusion_conditioning_latent]) A tuple of (autoregressive_conditioning_latent, diffusion_conditioning_latent), which can be provided in lieu of voice_samples. This is ignored unless `voice_samples=None`. Conditioning latents can be retrieved via `get_conditioning_latents()`. k: (int) The number of returned clips. The most likely (as determined by Tortoises' CLVP model) clips are returned. latent_averaging_mode: (int) 0/1/2 for following modes: 0 - latents will be generated as in original tortoise, using ~4.27s from each voice sample, averaging latent across all samples 1 - latents will be generated using (almost) entire voice samples, averaged across all the ~4.27s chunks 2 - latents will be generated using (almost) entire voice samples, averaged per voice sample verbose: (bool) Whether or not to print log messages indicating the progress of creating a clip. Default=true. num_autoregressive_samples: (int) Number of samples taken from the autoregressive model, all of which are filtered using CLVP. As Tortoise is a probabilistic model, more samples means a higher probability of creating something "great". temperature: (float) The softmax temperature of the autoregressive model. length_penalty: (float) A length penalty applied to the autoregressive decoder. Higher settings causes the model to produce more terse outputs. repetition_penalty: (float) A penalty that prevents the autoregressive decoder from repeating itself during decoding. Can be used to reduce the incidence of long silences or "uhhhhhhs", etc. top_p: (float) P value used in nucleus sampling. (0,1]. Lower values mean the decoder produces more "likely" (aka boring) outputs. max_mel_tokens: (int) Restricts the output length. (0,600] integer. Each unit is 1/20 of a second. typical_sampling: (bool) Turns typical sampling on or off. This sampling mode is discussed in this paper: https://arxiv.org/abs/2202.00666 I was interested in the premise, but the results were not as good as I was hoping. This is off by default, but could use some tuning. typical_mass: (float) The typical_mass parameter from the typical_sampling algorithm. diffusion_iterations: (int) Number of diffusion steps to perform. [0,4000]. More steps means the network has more chances to iteratively refine the output, which should theoretically mean a higher quality output. Generally a value above 250 is not noticeably better, however. cond_free: (bool) Whether or not to perform conditioning-free diffusion. Conditioning-free diffusion performs two forward passes for each diffusion step: one with the outputs of the autoregressive model and one with no conditioning priors. The output of the two is blended according to the cond_free_k value below. Conditioning-free diffusion is the real deal, and dramatically improves realism. cond_free_k: (float) Knob that determines how to balance the conditioning free signal with the conditioning-present signal. [0,inf]. As cond_free_k increases, the output becomes dominated by the conditioning-free signal. diffusion_temperature: (float) Controls the variance of the noise fed into the diffusion model. [0,1]. Values at 0 are the "mean" prediction of the diffusion network and will sound bland and smeared. hf_generate_kwargs: (**kwargs) The huggingface Transformers generate API is used for the autoregressive transformer. Extra keyword args fed to this function get forwarded directly to that API. Documentation here: https://huggingface.co/docs/transformers/internal/generation_utils Returns: Generated audio clip(s) as a torch tensor. Shape 1,S if k=1 else, (k,1,S) where S is the sample length. Sample rate is 24kHz. r'r)rr+rrLzVToo much text provided. Break the text up into separate segments and re-try inference.NT)r'r(r))r:r;r<r8rvrMz#Generating autoregressive samples..r?) device_typedtypeenabled)disable) do_samplerEronum_return_sequencesrCrDmax_generate_lengthr+)r F) return_lossr)krb) return_latent clip_inputsz0Transforming autoregressive outputs into audio..r)rorpcs$jrj|d|dS|S)Nr+)rrredactrCrB)rEr;rrr potentially_redact&sz.Tortoise.inference..potentially_redactcsg|]}|qSrr)r wav_candidate)rkr;rr r+sz&Tortoise.inference..)r9r:r;r&rn)(r(r# IntTensorrencoderBrDrGrrrr5r8r=rrgrstop_mel_tokenrUrautocastfloat16rr$inference_speechrr r]repeatcattopkindicesr7mel_length_compressionrrurrXrrT)7rr;r&rnrgrpuse_deterministic_seedreturn_deterministic_stater(rGrorCrDrErrHr;r<rFr8halfr)hf_generate_kwargsr: text_tokensauto_conditioningdiffusion_conditioningrrlorig_batch_sizesamples num_batchesro calm_tokenrbrXpadding_needed clip_resultsr batchiclvp_res best_results best_latentswav_candidatesrmctokenscoderrtrr9resrAr)rkrr;r rXVs4 H                  ozTortoise.inferencecCtdNz$Tortoise Training is not implementedNotImplementedErrorrrrr forwardCzTortoise.forwardcCrrrrrrr eval_stepFrzTortoise.eval_stepTortoiseConfigcKst|Sr)r)rrHrrr init_from_configIszTortoise.init_from_configc  KsT|jdur||_|ptj|d} |ptj|d} |p"tj|d} |p+tj|d}tj|d|_tj| rLtj| tdd} |j j | d d tj| r]|j j t| |d tj| rn|j j t| |d tj|r|j |jj jtj|tdd|r|j |jj|j |j |j |j dSdS) aLoad a model checkpoints from a directory. This model is with multiple checkpoint files and it expects to have all the files to be under the given `checkpoint_dir` with the rigth names. If eval is True, set the model to eval mode. Args: config (TortoiseConfig): The model config. checkpoint_dir (str): The directory where the checkpoints are stored. ar_checkpoint_path (str, optional): The path to the autoregressive checkpoint. Defaults to None. diff_checkpoint_path (str, optional): The path to the diffusion checkpoint. Defaults to None. clvp_checkpoint_path (str, optional): The path to the CLVP checkpoint. Defaults to None. vocoder_checkpoint_path (str, optional): The path to the vocoder checkpoint. Defaults to None. eval (bool, optional): Whether to set the model to eval mode. Defaults to False. strict (bool, optional): Whether to load the model strictly. Defaults to True. Nzautoregressive.pthzdiffusion_decoder.pthz clvp2.pthz vocoder.pthz mel_norms.pthrrF)strict)rrrrrexistsr#rrGrrrr r model_argsr optionally_indexpost_init_gpt2_configrrr6)rrcheckpoint_dirar_checkpoint_pathdiff_checkpoint_pathclvp_checkpoint_pathvocoder_checkpoint_pathr6rrHar_path diff_path clvp_path checkpointrrr load_checkpointMs<          zTortoise.load_checkpointcCrrrrrrr train_steprzTortoise.train_step)FrF)r%N)NNr+TNFrrr[r+r\r[r]raTrvr^r,TF)rr)NNNNFT)rrrrrrrrr5r8rBr=rXrr staticmethodrrr __classcell__rrrr r5s\ J   K (= n  Arr)r)r*Tr+r,)r>r?)T)r+T)8rr% contextlibr dataclassesrrr#torch.nn.functionalnnr!rr coqpitrr"TTS.tts.layers.tortoise.arch_utilsr#TTS.tts.layers.tortoise.audio_utilsrr r &TTS.tts.layers.tortoise.autoregressiver "TTS.tts.layers.tortoise.classifierr TTS.tts.layers.tortoise.clvpr !TTS.tts.layers.tortoise.diffusionrrr)TTS.tts.layers.tortoise.diffusion_decoderr/TTS.tts.layers.tortoise.random_latent_generatorr!TTS.tts.layers.tortoise.tokenizerrTTS.tts.layers.tortoise.vocoderrr)TTS.tts.layers.tortoise.wav2vec_alignmentrTTS.tts.models.base_ttsrr!r(r=rLr]rurrrrrrrrr sT                 # k