P/j< ddlZddlmZddlmZmZmZmZmZm Z ddl Z ddl m Z ddl m Z ddlmZddlmZmZddlmZmZd d lmZd d lmZgZd ZGd dej8ZGddej8ZGdde j>j@ejBZ"Gdde j>j@ejBZ#GddZ$GddZ%eGddZ&eGddZ'GddZ(eGdd e'e&e$eZ)eGd!d"e'e&e%eZ*eGd#d$e(e&e$eZ+eGd%d&e(e&e%eZ,e+d'ejZd()*Z.d+e._/e,d,ejZd-)*Z0d.e0_/e)d/ejZd()d0ejb1Z2d2e2_/e*d3ejZd-)d0ejb1Z3d4e3_/y)5N) dataclass)AnyDictListOptionalTupleUnion)Tensor)load_state_dict_from_url)mu_law_decoding) Tacotron2WaveRNN) GriffinLimInverseMelScale)utils)Tacotron2TTSBundlez.https://download.pytorch.org/torchaudio/modelscXeZdZfdZedZdeeeefde e e ffdZ xZ S)_EnglishCharProcessorct|tj|_t |jDcic]\}}|| c}}|_ycc}}wN)super__init__r _get_chars_tokens enumerate_mapping)selfis __class__s a/root/repos/pocs/video-evals/.venv/lib/python3.12/site-packages/torchaudio/pipelines/_tts/impl.pyrz_EnglishCharProcessor.__init__sD '') *3DLL*AB$!QAB Bs Ac|jSrrrs r"tokensz_EnglishCharProcessor.tokens ||textsreturnc t|tr|g}|Dcgc]<}|jDcgc] }||jvs|j|"c}>}}}t j |Scc}wcc}}wr) isinstancestrlowerrr _to_tensor)rr)tcindicess r"__call__z_EnglishCharProcessor.__call__#sf eS !GEX]^STaggiN1 ;MDMM!$N^^((O^sA7A2 A2 A72A7 __name__ __module__ __qualname__rpropertyr&r r-rrr r3 __classcell__r!s@r"rrsFC )eCcN3)ffn8M)r(rc^eZdZddfd ZedZdeeeefde e e ffdZ xZ S)_EnglishPhoneProcessorN dl_kwargsct|tj|_t |jDcic]\}}|| c}}|_tjd||_d|_ ycc}}w)Nzen_us_cmudict_forward.ptr=z(\[[A-Z]+?\]|[_!'(),.:;? -])) rrr _get_phonesrrr_load_phonemizer _phonemizer_pattern)rr>rpr!s r"rz_EnglishPhoneProcessor.__init__+se ((* *3DLL*AB$!QAB  112LXab7 Cs A;c|jSrr$r%s r"r&z_EnglishPhoneProcessor.tokens2r'r(r)r*c vt|tr|g}g}|j|dD]p}tj|j |Dcgc]}tj dd|}}|j|Dcgc]}|j|c}rtj|Scc}wcc}w)Nen_us)langz[\[\]]) r,r-rBrefindallrCsubappendrrr/)rr)r2phonesrretrDs r"r3z_EnglishPhoneProcessor.__call__6s eS !GE&&u7&; )minrr)r`max) torchexplog10clamprYrXinferr_unnormalize_waveformn_bitsr n_classessqueeze)rmel_speclengthswaveforms r"forwardz_WaveRNNVocoder.forwardRs99X& EKKd$CDD    )**X59K9KKH{{8:H KK--h@'..x9K9KL"8T[[-B-BC##A&  r()ir) r5r6r7rrfloatrr8r\rnr9r:s@r"rRrRGs3*g*Xe_* !! !r(rRc6eZdZfdZedZddZxZS)_GriffinLimVocoderc t|d|_tdd|jdddd|_t dd d d |_y) NrViPgg@@slaney)n_stftn_melsr\f_minf_max mel_scalenormir)n_fftpower hop_length win_length)rrrWrr\_inv_melr _griffin_lim)rr!s r"rz_GriffinLimVocoder.__init__`sX !'!((  '  r(c|jSrr[r%s r"r\z_GriffinLimVocoder.sample_ratesr]r(ctj|}|jjj d}|j |}|jj d}|j |}||fS)NTF)rbrcclonedetachrequires_grad_rr)rrkrlspec waveformss r"rnz_GriffinLimVocoder.forwardwso99X&>>#**,;;DA}}X&{{}++E2%%d+ '!!r(r)r5r6r7rr8r\rnr9r:s@r"rqrq_s! &!!"r(rqc,eZdZdejfdZy) _CharMixinr*ctSr)rr%s r"get_text_processorz_CharMixin.get_text_processors $&&r(Nr5r6r7r TextProcessorrr(r"rrs'$6$D$D'r(rc2eZdZdddejfdZy) _PhoneMixinNr=r*ct|SNr=)r<)rr>s r"rz_PhoneMixin.get_text_processors % ::r(rrr(r"rrs.2;7I7W7W;r(rc>eZdZUeed<eeefed<dddefdZy)_Tacotron2Mixin_tacotron2_path_tacotron2_paramsNr=r*ctdi|j}td|j}|in|}t |fi|}|j ||j |SN/r)r r _BASE_URLrr load_state_dictevalrr>rSurl state_dicts r" get_tacotron2z_Tacotron2Mixin.get_tacotron2sc3D223 1T1123#+B -c?Y?  j)  r() r5r6r7r-__annotations__rrr rrr(r"rrs%CH~%)-)r(rcPeZdZUeeed<eeeefed<dddZdddZ y) _WaveRNNMixin _wavernn_path_wavernn_paramsNr=c<|j|}t|Sr) _get_wavernnrR)rr>wavernns r" get_vocoderz_WaveRNNMixin.get_vocoders ##i#8w''r(ctdi|j}td|j}|in|}t |fi|}|j ||j |Sr)rrrrr rrrs r"rz_WaveRNNMixin._get_wavernnsc/$../ 1T//01#+B -c?Y?  j)  r() r5r6r7rr-rrrrrrr(r"rrs1C= d38n--'+()-r(rceZdZdZy)_GriffinLimMixinc tSr)rq)r_s r"rz_GriffinLimMixin.get_vocoders !##r(N)r5r6r7rrr(r"rrs$r(rc eZdZy)_Tacotron2WaveRNNCharBundleNr5r6r7rr(r"rrr(rc eZdZy)_Tacotron2WaveRNNPhoneBundleNrrr(r"rrrr(rc eZdZy)_Tacotron2GriffinLimCharBundleNrrr(r"rrrr(rc eZdZy)_Tacotron2GriffinLimPhoneBundleNrrr(r"rrrr(rz5tacotron2_english_characters_1500_epochs_ljspeech.pth&) n_symbols)rraCharacter-based TTS pipeline with :py:class:`~torchaudio.models.Tacotron2` trained on *LJSpeech* :cite:`ljspeech17` for 1,500 epochs, and :py:class:`~torchaudio.transforms.GriffinLim` as vocoder. The text processor encodes the input texts character-by-character. You can find the training script `here `__. The default parameters were used. Please refer to :func:`torchaudio.pipelines.Tacotron2TTSBundle` for the usage. Example - "Hello world! T T S stands for Text to Speech!" .. image:: https://download.pytorch.org/torchaudio/doc-assets/TACOTRON2_GRIFFINLIM_CHAR_LJSPEECH.png :alt: Spectrogram generated by Tacotron2 .. raw:: html Example - "The examination and testimony of the experts enabled the Commission to conclude that five shots may have been fired," .. image:: https://download.pytorch.org/torchaudio/doc-assets/TACOTRON2_GRIFFINLIM_CHAR_LJSPEECH_v2.png :alt: Spectrogram generated by Tacotron2 .. raw:: html z3tacotron2_english_phonemes_1500_epochs_ljspeech.pth`aPhoneme-based TTS pipeline with :py:class:`~torchaudio.models.Tacotron2` trained on *LJSpeech* :cite:`ljspeech17` for 1,500 epochs and :py:class:`~torchaudio.transforms.GriffinLim` as vocoder. The text processor encodes the input texts based on phoneme. It uses `DeepPhonemizer `__ to convert graphemes to phonemes. The model (*en_us_cmudict_forward*) was trained on `CMUDict `__. You can find the training script `here `__. The text processor is set to the *"english_phonemes"*. Please refer to :func:`torchaudio.pipelines.Tacotron2TTSBundle` for the usage. Example - "Hello world! T T S stands for Text to Speech!" .. image:: https://download.pytorch.org/torchaudio/doc-assets/TACOTRON2_GRIFFINLIM_PHONE_LJSPEECH.png :alt: Spectrogram generated by Tacotron2 .. raw:: html Example - "The examination and testimony of the experts enabled the Commission to conclude that five shots may have been fired," .. image:: https://download.pytorch.org/torchaudio/doc-assets/TACOTRON2_GRIFFINLIM_PHONE_LJSPEECH_v2.png :alt: Spectrogram generated by Tacotron2 .. raw:: html z=tacotron2_english_characters_1500_epochs_wavernn_ljspeech.pthz%wavernn_10k_epochs_8bits_ljspeech.pth)rrrraCharacter-based TTS pipeline with :py:class:`~torchaudio.models.Tacotron2` trained on *LJSpeech* :cite:`ljspeech17` for 1,500 epochs and :py:class:`~torchaudio.models.WaveRNN` vocoder trained on 8 bits depth waveform of *LJSpeech* :cite:`ljspeech17` for 10,000 epochs. The text processor encodes the input texts character-by-character. You can find the training script `here `__. The following parameters were used; ``win_length=1100``, ``hop_length=275``, ``n_fft=2048``, ``mel_fmin=40``, and ``mel_fmax=11025``. You can find the training script `here `__. Please refer to :func:`torchaudio.pipelines.Tacotron2TTSBundle` for the usage. Example - "Hello world! T T S stands for Text to Speech!" .. image:: https://download.pytorch.org/torchaudio/doc-assets/TACOTRON2_WAVERNN_CHAR_LJSPEECH.png :alt: Spectrogram generated by Tacotron2 .. raw:: html Example - "The examination and testimony of the experts enabled the Commission to conclude that five shots may have been fired," .. image:: https://download.pytorch.org/torchaudio/doc-assets/TACOTRON2_WAVERNN_CHAR_LJSPEECH_v2.png :alt: Spectrogram generated by Tacotron2 .. raw:: html z;tacotron2_english_phonemes_1500_epochs_wavernn_ljspeech.pthaPhoneme-based TTS pipeline with :py:class:`~torchaudio.models.Tacotron2` trained on *LJSpeech* :cite:`ljspeech17` for 1,500 epochs, and :py:class:`~torchaudio.models.WaveRNN` vocoder trained on 8 bits depth waveform of *LJSpeech* :cite:`ljspeech17` for 10,000 epochs. The text processor encodes the input texts based on phoneme. It uses `DeepPhonemizer `__ to convert graphemes to phonemes. The model (*en_us_cmudict_forward*) was trained on `CMUDict `__. You can find the training script for Tacotron2 `here `__. The following parameters were used; ``win_length=1100``, ``hop_length=275``, ``n_fft=2048``, ``mel_fmin=40``, and ``mel_fmax=11025``. You can find the training script for WaveRNN `here `__. Please refer to :func:`torchaudio.pipelines.Tacotron2TTSBundle` for the usage. Example - "Hello world! T T S stands for Text to Speech!" .. image:: https://download.pytorch.org/torchaudio/doc-assets/TACOTRON2_WAVERNN_PHONE_LJSPEECH.png :alt: Spectrogram generated by Tacotron2 .. raw:: html Example - "The examination and testimony of the experts enabled the Commission to conclude that five shots may have been fired," .. image:: https://download.pytorch.org/torchaudio/doc-assets/TACOTRON2_WAVERNN_PHONE_LJSPEECH_v2.png :alt: Spectrogram generated by Tacotron2 .. raw:: html )4rJ dataclassesrtypingrrrrrr rbr torchaudio._internalr torchaudio.functionalr torchaudio.modelsr rtorchaudio.transformsrrrIr interfacer__all__rrrr<nnModuleVocoderrRrqrrrrrrrrr_get_taco_params"TACOTRON2_GRIFFINLIM_CHAR_LJSPEECH__doc__#TACOTRON2_GRIFFINLIM_PHONE_LJSPEECH_get_wrnn_paramsTACOTRON2_WAVERNN_CHAR_LJSPEECH TACOTRON2_WAVERNN_PHONE_LJSPEECHrr(r"rsB !:: 910=)  < ).<<)")/==):!ehhoo'9'A'A!0"*<*D*D"L'' ;;        $$$  -*N`     =/;Pb     %5 Tf     &6Vh   &DK,e,,r:&"!."*F'FI,e,,r:'#&/#+P#>S,e,,r:9*E**, # #+'J$@Q,e,,r:9*E**, $ ), (r(