o j+@sddlmZddlZddlZddlmZd"ddZd"dd Z     d#d ejd ej d e d ej de dej dej defddZ ddZ ddZd"ddZd"ddZddZ       d$ddZ      d%d d!ZdS)&)DictN)nnFcpucCs(|rd}|dur dStj|||d}|S)Ncuda)dtypedevice)torch as_tensor)np_arrayrrrtensorr J/home/kuhnn/.local/lib/python3.10/site-packages/TTS/tts/utils/synthesis.pynumpy_to_torchs rcCs2|rd}tj||j||jd|dd}|S)Nrsrrr)r FloatTensormelspectrogramload_wav sample_rate unsqueeze) style_wavaprr style_melr r r compute_style_melsrmodelinputs speaker_idr style_textd_vector language_idreturnc CsTt|jdd|j}t|dr|jj}n|j}||||||||dd} | S)a!Run a torch model for inference. It does not support batch inference. Args: model (nn.Module): The model to run inference. inputs (torch.Tensor): Input tensor with character ids. speaker_id (int, optional): Input speaker ids for multi-speaker models. Defaults to None. style_mel (torch.Tensor, optional): Spectrograms used for voice styling . Defaults to None. d_vector (torch.Tensor, optional): d-vector for multi-speaker models . Defaults to None. Returns: Dict: model outputs. module) x_lengths speaker_ids d_vectorsrr language_ids) aux_input)rr shapetorhasattrr$ inference) rrrrrrr input_lengths_funcoutputsr r r run_model_torchs   r1cCs|d||S)N) find_endpoint)wavrr r r trim_silenceCsr4cCs.|jdvr||j}|S||j}|S)N)tacotron)rlowerinv_spectrogramTinv_melspectrogram)postnet_outputrCONFIGr3r r r r7Gs   r7cCs.|rd}|durt|}t||}|S)Nr)npasarrayr from_numpyr+)aux_idrrr r r id_to_torchOs  r@cCsD|rd}|dur t|}t|tj}|d|}|S)Nrr) r<r=rr>typersqueezerr+)rrrr r r embedding_to_torchXs rCc CsLg}t|D]\}}|||j|j}t|||}||d|q|S)aYApply griffin-lim to each sample iterating throught the first dimension. Args: inputs (Tensor or np.Array): Features to be converted by GL. First dimension is the batch size. input_lens (Tensor or np.Array): 1D array of sample lengths. CONFIG (Dict): TTS config. ap (AudioProcessor): TTS audio processor. N) enumerate hop_lengthr7append) r input_lensr;rwavsidxspecwav_lenr3r r r apply_griffin_limcs  rLc  s t|j} |r d} d} |dr)|jr)|dur)t|tr!|} nt||j| d} |drC|j rC|durCt||j| d} | dd} d} durdfdd |j j D}t|dks`Jd |d } tj|jj|| d tjd }|dur|t|| d}| durt| | d} durt| dt| tst| tj| d} |durtj|jj|d tjd }t|tj| d}|d }t|tj| d}|d }t|||| || d}|d}|d j}|d}d}|}|j dkr|rt!||j|}|rt"||j}n|}||||d}|S)aSynthesize voice for the given text using Griffin-Lim vocoder or just compute output features to be passed to the vocoder model. Args: model (TTS.tts.models): The TTS model to synthesize audio with. text (str): The input text to convert to speech. CONFIG (Coqpit): Model configuration. use_cuda (bool): Enable/disable CUDA. speaker_id (int): Speaker ID passed to the speaker embedding layer in multi-speaker model. Defaults to None. style_wav (str | Dict[str, float]): Path or tensor to/of a waveform used for computing the style embedding based on GST or Capacitron. Defaults to None, meaning that Capacitron models will sample from the prior distribution to generate random but realistic prosody. style_text (str): Transcription of style_wav for Capacitron models. Defaults to None. enable_eos_bos_chars (bool): enable special chars for end of sentence and start of sentence. Defaults to False. do_trim_silence (bool): trim silence after synthesis. Defaults to False. d_vector (torch.Tensor): d-vector for multi-speaker models in share :math:`[1, D]`. Defaults to None. language_id (int): Language ID passed to the language embedding layer in multi-langual model. Defaults to None. rNgstrcapacitron_vaer"r#csg|] \}}|kr|qSr r ).0kvr r r szsynthesis..z$language_id must be a valid languager)language)r)rr model_outputs alignments)r3rV text_inputsr0)#next parametersrhasrM isinstancedictrruse_capacitron_vae transposelanguage_manager name_to_iditemslenr<r= tokenizer text_to_idsint32r@rCrrfloatlongrr1datarnumpyrBndimr7r4)rtextr;use_cudarrruse_griffin_limdo_trim_silencerr rr language_namerTrWr0rUrVr3 return_dictr rRr synthesists5           rqc Cst|j} |r d} |durt|| d}|durt|| d}|dur)t|| d}t|jj||jjr6|jjn|jj d| d}t |drI|j j } n|j } | |||||} d} | } | jdkrr| rpt| |j|} |rpt| |j} | S| } | S)aSynthesize voice for the given text using Griffin-Lim vocoder or just compute output features to be passed to the vocoder model. Args: model (TTS.tts.models): The TTS model to synthesize audio with. CONFIG (Coqpit): Model configuration. use_cuda (bool): Enable/disable CUDA. reference_wav (str): Path of reference_wav to be used to voice conversion. speaker_id (int): Speaker ID passed to the speaker embedding layer in multi-speaker model. Defaults to None. d_vector (torch.Tensor): d-vector for multi-speaker models in share :math:`[1, D]`. Defaults to None. reference_speaker_id (int): Reference Speaker ID passed to the speaker embedding layer in multi-speaker model. Defaults to None. reference_d_vector (torch.Tensor): Reference d-vector for multi-speaker models in share :math:`[1, D]`. Defaults to None. enable_eos_bos_chars (bool): enable special chars for end of sentence and start of sentence. Defaults to False. do_trim_silence (bool): trim silence after synthesis. Defaults to False. rNrrr$r#)rXrYrr@rCrrargsencoder_sample_raterr,r$inference_voice_conversionrBrjr7r4)rr;rl reference_wavrrreference_speaker_idreference_d_vectorrnrmrr/rUr3r r r transfer_voices:/       rx)Fr)NNNNN)NNNFFNN)NNNNFF)typingrrir<rrrrModuleTensorintstrr1r4r7r@rCrLrqrxr r r r sb      (