^j*ddlZddlmZddlmZmZddlmZddlm Z ddl m Z m Z m Z ddlmZmZer ddlZddlmcmZerddlZe j.eZGd d eZd gZy) N)Any) AudioInputmake_list_of_audio)SequenceFeatureExtractor) BatchFeature)PaddingStrategy TensorTypelogging)is_torch_availableis_torchaudio_availableceZdZdZgdZ dfd Z ddedeeze zde dzded ee zdzd e dzd ed e fd Z d eeeffdZxZS)Xcodec2FeatureExtractora Constructs a Xcodec2 feature extractor, which computes mel-filter bank features for the semantic encoder and padded audio for the acoustic encoder. This feature extractor inherits from [`SequenceFeatureExtractor`] which contains most of the main methods. Users should refer to this superclass for more information regarding those methods. Args: feature_size (`int`, *optional*, defaults to 80): The feature dimension of the extracted features. sampling_rate (`int`, *optional*, defaults to 16000): The sample rate at which the audio files should be digitalized expressed in hertz (Hz). padding_value (`float`, *optional*, defaults to 1.0): The value that is used to fill the padding vectors for the mel spectrogram. hop_length (`int`, *optional*, defaults to 320): Number of audio samples encoded per frame. Equivalent to product of downsampling ratios. Needed for acoustic encoder input padding. )input_features input_values padding_maskinput_features_maskc t|d |||d|||_td|d|_ddg|j_d|_d|_d|_d |_ y) N) feature_size sampling_rate padding_valueaudiorPi) super__init__ hop_lengthracoustic_encoder_paddermodel_input_namesstride num_mel_bins frame_length frame_shift)selfrrrr!kwargs __class__s /var/www/ramen.bs-engineer-server.com/venv/lib/python3.12/site-packages/transformers/models/xcodec2/feature_extraction_xcodec2.pyr z Xcodec2FeatureExtractor.__init__:sv wl-_lwpvw%'?'( $ ;B>9R$$6 Nrpadding max_length truncationreturn_tensorsrdevicereturnc `ts td|;||jk7rYtd|d|jd|jd|d tj d|j jd t|}|D])} | jd kDstd | jt|} |D cgc]-} tjtj| d d /}} |j jt#d|i|||||j$d} | j'd} | ddddddf}g}t)| D]}| *t+| |j-j/n|jd}||j$zdz |j$z|j$z}t1||jd}||ddd|f}tj||j$d z|j$d zfd }|j3|}t4j6j8j;|dz|j<|j>|jz dz|j@|jz dz|jdddddd dd|jd z}||jCdz tjD|jGdddzz }|jI|t#d |i}|j|||||jJ|d!}|d }|jMd}|jd|jd|jJzz }|ddd|ddfjO| ||jJz|j<|jJz}|M|ddd|fjO| ||jJz|jJj1d"jP}t#|| ||d#|$Scc} w)%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. z;PyTorch is required for mel-filter bank feature extraction.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.rz6Expected input shape (channels, length) but got shape )rrr)valuerpt)r.r/r-return_attention_maskpad_to_multiple_ofr0attention_maskriipoveyg ףp= ?TF) r%r&r'sample_frequency window_typepreemphasis_coefficientremove_dc_offset use_log_fbank use_energydither snip_edgeslow_freq high_freqr)unbiasedgHz>r)r-r.r/r8r7r0)dim)rrrr) tensor_type))r ImportErrorr ValueErrorloggerwarningr*__name__rndimshapelenFpadtorch as_tensorr"rr!poprangeintsumitemminto torchaudio compliancekaldifbankr%r&r'meansqrtvarappendr$getreshapevalues)r(rr-r.r/r0rr1r)example batch_sizea padded_inputsr padded_audio mel_featuresiorig_lenper_sample_len valid_lenwaveformfeaturesencoded_inputs padded_melaudio_spectrogramspectrogram_masktrimmed_framess r+__call__z Xcodec2FeatureExtractor.__call__TstZ"#[\ \  $ 2 22 I$P**+,**+9]O1F NNVW[WeWeWnWnVop\\  #5) kG||a #YZaZgZgYh!ijj kZ HMM!uq)6=MM4488 '5) *!!")#9 %(()9: $W-aqj9  z" *A,>r,BCI#Aq*9*$45HuuX1(R'S[^_H{{6*H!,,2288E"!..!..1C1CCdJ ,,t/A/AADH!%!3!3#(,!%" ,,19H !8==#33uzz(,,q[_,B`cgBg7hhH    )1 *2&'7&FGXX !!#{{") ''78%>>*:;*0037H7N7Nq7QTXT_T_7_`-a.!.CDLL $++5t7H7H4;;7V   ' O^O!34^t{{%BDKKP  , ,"3'7   '  ENs2P+ctj|j}|jj|d<|j dd|S)Nfeature_extractor_typer")copydeepcopy__dict__r*rNrV)r(outputs r+to_dictzXcodec2FeatureExtractor.to_dicts=t}}-+/>>+B+B'( ,d3 r,)ri>g?i@)TNFNNcpu)rN __module__ __qualname____doc__r#r rboolstrr rXr rrydictrr __classcell__)r*s@r+rr$s&b :15!% 26$(N N o-N $J N  N j(4/ N TzN N  N `c3hr,r)r|typingr audio_utilsrr!feature_extraction_sequence_utilsrfeature_extraction_utilsrutilsr r r utils.import_utilsr r rTtorch.nn.functionalnn functionalrRr] get_loggerrNrLr__all__rr,r+rsg 9I499M##   H %D6DN % %r,