+ jL ^RIt^RIHt^RIHtHt^RIHt^RI H t ^RI H t ^RI HtHtHtHtHtHtHt^RIHtHt^R IHt^R IHt^R IHtHtHtH t H!t!]!4'd^RI"t"^RI#H$uH%t&]!4'd^RI't']PP!])4t*]!R R 7]!RR]444t+!RR] 4t,!RR]4t-!RR]4t.!RR]!4t/]!RR7!RR]44t0!RR]4t1.ROt2R#)N)strict) AudioInputmake_list_of_audio) BatchFeature)create_bidirectional_mask)Unpack)PaddingStrategy TensorTypeTransformersKwargsauto_docstringcan_return_tupleis_torch_availablelogging)is_torchaudio_availablerequires) Xcodec2Config)Xcodec2FeatureExtractor)Xcodec2DecoderOutputXcodec2EncoderOutput Xcodec2Model Xcodec2OutputXcodec2PreTrainedModelzneuphonic/neucodec) checkpointca]tRt^6toRtRtRtRt]!4t ] V3RlRl4t ] V3RlRl4t V3R lt R tVtR #) NeuCodecConfigaq downsampling_ratios (`list[int]`, *optional*, defaults to `[2, 2, 4, 4, 5]`): Ratios for downsampling in the encoder. semantic_model_config (`Union[Dict, Wav2Vec2BertConfig]`, *optional*): An instance of the configuration object for the semantic (Wav2Vec2BertConfig) model. quantization_dim (`int`, *optional*, defaults to 2048): Dimension for the vector quantization codebook. quantization_levels (`list[int]`, *optional*, defaults to `[4, 4, 4, 4, 4, 4, 4, 4]`): Levels for the vector quantization codebook. input_sampling_rate (`int`, *optional*, defaults to 16000): Sampling rate, in hertz (Hz), of the encoder's input audio waveform. output_sampling_rate (`int`, *optional*, defaults to 24000): Sampling rate, in hertz (Hz), of the decoder's output audio waveform. Example: ```python >>> from transformers import NeuCodecConfig, NeuCodecModel >>> # Initializing configuration >>> configuration = NeuCodecConfig() >>> # Initializing a model (with random weights) from the configuration >>> model = NeuCodecModel(configuration) >>> # Accessing the model configuration >>> configuration = model.config ```neucodeci>i]c <V^8dQhRS[/#returnint)format __classdict__s"n/home/luca/screen-checking/.venv/lib/python3.14/site-packages/transformers/models/neucodec/modular_neucodec.py __annotate__NeuCodecConfig.__annotate__\s66C6c T\\P!VP44#)zhHop length, in samples at `input_sampling_rate`, between successive codes (i.e. the encoder frame rate).)r"npproddownsampling_ratiosselfs&r%encoder_hop_length!NeuCodecConfig.encoder_hop_length[s277433455r(c <V^8dQhRS[/#rr!)r#r$s"r%r&r'asccCcr(cp\VPVP,VP, 4#N)r"r/output_sampling_rateinput_sampling_rater-s&r% hop_lengthNeuCodecConfig.hop_length`s+4**T-F-FFIaIaabbr(c2<V^8dQh/S[;R&S[;R&#)rr5r4r!)r#r$s"r%r&r'6s#B%CD&Er(N)__name__ __module__ __qualname____firstlineno____doc__ model_typer5r4AttributeError sampling_ratepropertyr/r6__annotate_func____static_attributes____classdictcell__r$s@r%rr6sU :J% &"$M 66ccWr(rc]tRt^gtRtR#)NeuCodecOutputr9Nr:r;r<r=rDr9r(r%rHrHgr(rHc]tRt^ktRtR#)NeuCodecEncoderOutputr9NrIr9r(r%rLrLkrJr(rLc]tRt^otRtR#)NeuCodecDecoderOutputr9NrIr9r(r%rNrNorJr(rNc]tRt^stRtR#)NeuCodecPreTrainedModelr9NrIr9r(r%rPrPsrJr(rPz"NeuCodec neural audio codec model.) custom_introcaa]tRt^wtoV3RlV3Rllt]]R V3RlRll44t]]R V3RlRll44t]]R V3RlRll44t R t Vt V;t #) NeuCodecModelc <V^8dQhRS[/#)rconfig)r)r#r$s"r%r&NeuCodecModel.__annotate__ysff~fr(c<\SV`V4VPVnVPVP , VnR#r3)super__init__r/r6r4r5sample_rate_conversion_factor)r.rU __class__s&&r%rYNeuCodecModel.__init__ys8  !33-3-H-H6KeKe-e*r(c<V^8dQhRS[PRS[PRS[PR,RS[PR,RS[RS[S[,RS[S[,/# r input_valuesinput_features padding_maskNinput_features_maskoutput_latentskwargsr )torchTensorboolrr tuplerL)r#r$s"r%r&rVsw: : ll:  : llT) : #\\D0 :  : +,:  & &: r(c \P!4;_uu_4VPW$R7pRRR4XPP ^^4pVP V4pVP V4p \V PR ,VPR ,4p V RRV 13,p VRRV 13,p\P!W.^R7p VPV P ^^44p VPV 4wrV P ^^4p V P ^^4p RpVeVPR RR7pWP,pWP,p\P!V PR ,VP R7P#^R 4pVV8P%VP&4p\)T V'd V VR7#RVR7# +'giEL;i) a input_values (`torch.Tensor` of shape `(batch_size, 1, sequence_length)`): Input audio waveform. input_features (`torch.Tensor` of shape `(batch_size, mel_bins, time_steps)`): Input audio mel spectrogram for semantic encoding. padding_mask (`torch.Tensor` of shape `(batch_size, 1, sequence_length)`): Padding mask used to pad `input_values`. input_features_mask (`torch.Tensor` of shape `(batch_size, time_steps)`, *optional*): Attention mask for the spectrogram input to the semantic encoder. `1` for valid frames, `0` for padding. output_latents (`bool`, *optional*, defaults to `False`): Whether to return the continuous latent representation from the quantizer. )attention_maskN.dimT)rlkeepdim)device) audio_codeslatentsaudio_codes_mask)reno_gradsemantic_encoderlast_hidden_state transposesemantic_adapteracoustic_encoderminshapecat fc_encoder quantizersumrZr6arangernviewtodtyperL)r.r_r`rarbrcrdsemantic_outputsemantic_hidden_statesacoustic_hidden_states min_length hidden_statesrprorq audio_length token_lengthidxs&&&&&&, r%encodeNeuCodecModel.encodes2]]__"33N3gO!0!B!B!L!LQPQ!R!%!6!67M!N"&!6!6|!D/55b9;Q;W;WXZ;[\ !7[j[8H!I!7[j[8H!I #9"RXYZ  (?(?1(EF  $~~m<##Aq)!++Aq1    #'++D+AL'*L*LLL'??:L,,{004\=P=PQVVWXZ\]C #l 266|7I7IJ $#-G-  37-  ?__s G,, G= c <V^8dQhRS[PR,RS[PR,RS[PR,RS[S[,RS[S[,/#)rroNrprqrdr )rerfrr rhrN)r#r$s"r%r&rVsd ? ?\\D( ?$ ? ,,- ? +, ? & & ?r(c VfVf \R4hVe-VPPVP^^44pMVP^^4p\ VP VVR7pVP !V3RV/VBp\VR7#)a audio_codes (`torch.LongTensor` of shape `(batch_size, 1, codes_length)`): Discrete code indices computed using `model.encode`. latents (torch.Tensor of shape `(batch_size, dimension, time_steps)`, *optional*): Quantized continuous representation of input. audio_codes_mask (`torch.Tensor` of shape `(batch_size, codes_length)`, *optional*): Mask marking valid (non-padding) positions, as returned by `model.encode`. Needed for correct batched decoding of variable-length audio. z3Either `latents` or `audio_codes` must be provided.)rU inputs_embedsrjrj) audio_values) ValueErrorr} from_codesrvrrUacoustic_decoderrN)r.rorprqrdrj recon_audios&&&&, r%decodeNeuCodecModel.decodes$ ?{2RS S  "nn// 0E0Ea0KLG''1-G3;;!+  ++G]N]V\] $+>>r(c<V^8dQhRS[PRS[PRS[PR,RS[PR,RS[RS[S[,RS[S[,/#r^)rerfrgrr rhrH)r#r$s"r%r&rVs@ @ ll@  @ llT) @ #\\D0 @  @ +,@   @ r(c vVPR ,p\WpP,4pVPVVVVRRR7p VP!R RV P RV P RR/VB^,RRV13,p \T V PV'd V P MRV P R7#) a input_values (`torch.Tensor` of shape `(batch_size, 1, sequence_length)`): Input audio waveform, sampled at `config.input_sampling_rate`. input_features (`torch.Tensor` of shape `(batch_size, mel_bins, time_steps)`): Input audio mel spectrogram for semantic encoding. padding_mask (`torch.Tensor` of shape `(batch_size, 1, sequence_length)`): Padding mask used to pad `input_values`. input_features_mask (`torch.Tensor` of shape `(batch_size, time_steps)`, *optional*): Attention mask for the spectrogram input to the semantic encoder. `1` for valid frames, `0` for padding. output_latents (`bool`, *optional*, defaults to `False`): Whether to return the continuous latent representation from the quantizer. Examples: ```python >>> from datasets import load_dataset >>> from transformers import AutoFeatureExtractor, NeuCodecModel >>> dataset = load_dataset("hf-internal-testing/librispeech_asr_dummy", "clean", split="validation") >>> audio = dataset["train"]["audio"][0]["array"] >>> model_id = "neuphonic/neucodec" >>> model = NeuCodecModel.from_pretrained(model_id) >>> feature_extractor = AutoFeatureExtractor.from_pretrained(model_id) >>> inputs = feature_extractor(audio=audio, sampling_rate=feature_extractor.sampling_rate, return_tensors="pt") >>> outputs = model(**inputs) >>> audio_codes = outputs.audio_codes >>> audio_values = outputs.audio_values # sampled at 24kHz ```T)r`rarbrc return_dictrprqr.N)rrorprqrrr9) rzr"rZrrrprqrHro) r.r_r`rarbrcrd input_length output_lengthencoder_outputsrs &&&&&&, r%forwardNeuCodecModel.forwardsV$))"- L+M+MMN ++ )% 3 & {{ #++ ,==     .=. " %'33/=O++4,==   r()r6rZ)NNF)NNN) r:r;r<r=rYr r rrrrDrE __classcell__)r[r$s@@r%rSrSwsff: : : x ? ? ?D@ @ @ @ r(rScNa]tRtRto]!RR7RV3RlRll4tRtVtR#) NeuCodecFeatureExtractori()backendsNc<V^8dQhRS[RS[S[,S[,RS[R,RS[RS[S[,R,RS[R,RS[R S[/#) raudiopadding max_lengthN truncationreturn_tensorsrArnr )rrgstrr r"r r)r#r$s"r%r&%NeuCodecFeatureExtractor.__annotate__*sN N N o-N $J N  N j(4/ N TzN N  N r(c Ve>W`P8wd-\RV RVP RVP RV R2 4hM-\PRVPP R24\ V4pVF,p V P^8gK\R V P 24h \V4p VU u.uF0p \P!\P!V 4R!R R 7NK2 pp VPP\R V/4VVVVVP R R7p V P#R4p V R ,R",p.p\%V 4EFZpVPP\R VV,./4VP R R7pVR ,P'V4p\(P*P,P/VR#,VP0VP2VP, R,VP4VP, R,VPRRRRRR R^VP^,R7pVVP7^4, \P8!VP;^RR7R,4, pVP=V4EK] \RV/4pVPVVVVVP>VR R7pVR,pVPAR4pVP^,VP^,VP>,, pVRRV1R3,PCV VVP>,VP0VP>,4pVeTVRRV13,PCV VVP>,VP>4PER$R7PFp\RVRV RVRV/VR 7#uup i)%a Args: audio (`np.ndarray`, `torch.Tensor`, `list[np.ndarray]`, `list[torch.Tensor]`): Numpy array or torch tensor with shape (num_channels, sequence_length). A list of such arrays or tensors can also be provided for a batch of inputs. padding (`bool`, `str` or [`~utils.PaddingStrategy`], *optional*, defaults to `True`): Select a strategy to pad the returned sequences (according to the model's padding side and padding index) among: - `True` or `'longest'`: Pad to the longest sequence in the batch (or no padding if only a single sequence if provided). - `'max_length'`: Pad to a maximum length specified with the argument `max_length` or to the maximum acceptable input length for the model if that argument is not provided. - `False` or `'do_not_pad'` (default): No padding (i.e., can output a batch with sequences of different lengths). max_length (`int`, *optional*): Maximum length of the returned list and optionally padding length (see above). truncation (`bool`): Activates truncation to cut input sequences longer than *max_length* to *max_length*. return_tensors (`str` or [`~utils.TensorType`], *optional*): If set, will return tensors instead of list of python integers. Acceptable values are: - `'tf'`: Return TensorFlow `tf.constant` objects. - `'pt'`: Return PyTorch `torch.Tensor` objects. - `'np'`: Return Numpy `np.ndarray` objects. sampling_rate (`int`, *optional*): The sample rate at which the `audio` input was sampled. It is strongly recommended to pass `sampling_rate` at the forward call to prevent silent errors. device (`str`, *optional*, defaults to `"cpu"`): Device for PyTorch tensors during mel-filter bank feature extraction. kwargs (*optional*): Remaining dictionary of keyword arguments that will be passed to the tokenizer or the feature extractor. Nz3The model corresponding to this feature extractor: z& was trained using a sampling rate of zD. Please make sure that the provided `audio` input was sampled with z and not .zDIt is strongly recommended to pass the `sampling_rate` argument to `zN()`. Failing to do so can result in silent errors that might be hard to debug.z6Expected input shape (channels, length) but got shape g)valuerpt)rrrreturn_attention_maskpad_to_multiple_ofrrjNNN)rripoveyg ףp= ?TF) num_mel_bins frame_length frame_shiftsample_frequency window_typepreemphasis_coefficientremove_dc_offset use_log_fbank use_energydither snip_edgeslow_freq high_freq)unbiasedgHz>r`)rrrrrrrkr_rarb) tensor_type)r)rNrirr)$rArloggerwarningr[r:rndimrzlenFpadre as_tensoracoustic_encoder_padderrr6popranger torchaudio compliancekaldifbankrrrmeansqrtvarappendstridegetreshaperyvalues)r.rrrrrrArnrdexample batch_sizea padded_inputsra padded_audio mel_featuresisingle_padded_audiowaveformfeaturesencoded_inputs padded_melaudio_spectrogramspectrogram_masktrimmed_framess&&&&&&&&, r%__call__!NeuCodecFeatureExtractor.__call__)s^  $ 2 22 I$P**+,**+9]O1F3 NNVW[WeWeWnWnVop\\  #5)G||a #YZaZgZgYh!ijjZ HMMu!uq)6=uM4488 '5) *!!")#9 %(()9: $W-j9  z"A"&">">"B"Bgaz23#'??##C#  +7366v>H!,,2288E"!..!..1C1CCdJ ,,t/A/AADH!%!3!3#(,!%" ,,19H !8==#33uzz(,,q[_,B`cgBg7hhH    )1#2&'7&FGXX !!#{{") ''78%>>*:;*0037H7N7Nq7QTXT_T_7_`-a.!.CDLL $++5t7H7H4;;7V   ' O^O!34^t{{%BDKKP    "3%'7   '  INs;6O3r9)rer)TNFNNcpu)r:r;r<r=rrrDrErFs@r%rr(s% ./N N 0N r(r)rrSrPr)3numpyr*huggingface_hub.dataclassesr audio_utilsrrfeature_extraction_utilsr masking_utilsrprocessing_utilsrutilsr r r r r rrutils.import_utilsrrxcodec2.configuration_xcodec2r"xcodec2.feature_extraction_xcodec2rxcodec2.modeling_xcodec2rrrrrretorch.nn.functionalnn functionalrr get_loggerr:rrrHrLrNrPrSr__all__r9r(r%rs.946&D9H##   H %/0,c],c1,c^ ]  0  0  4 ABm Lm Cm `P 6P f er(