^j". ddlZddlZddlZddlmcmZddlm Z ddl m Z ddl m Z ddlmZmZmZddlmZej(eZdded ed ed efd Zed Gdde ZdgZy)N)mel_filter_bank)SequenceFeatureExtractor) BatchFeature)PaddingStrategy TensorTypelogging)requiresvaluename tolerancereturncpt|}t||z |kDrt|d|t|S)Nz. must resolve to an integer sample count, got )roundabs ValueErrorint)r r r roundeds /var/www/ramen.bs-engineer-server.com/venv/lib/python3.12/site-packages/transformers/models/inkling/feature_extraction_inkling.py _to_exact_intrs>ElG 57?i'D6!OPUwWXX w<)torch)backendsc`eZdZdZddgZ ddedededed ed edzf fd Zdd ejd e dejfdZ dde jeezee jzeeezdedzdee zezdedzdededzdedzde ezdzd e dzdefdZxZS)InklingFeatureExtractora Constructs a TML audio feature extractor, which converts raw audio waveforms into log-mel spectrogram features (mel filterbank energies in log10 space). The quantization of these features into discrete dMel bins is performed downstream by [`InklingProcessor`]. This feature extractor inherits from [`~feature_extraction_sequence_utils.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, i.e. the number of mel filterbanks. sampling_rate (`int`, *optional*, defaults to 16000): The sampling rate at which the audio files should be digitized, expressed in hertz (Hz). padding_value (`float`, *optional*, defaults to 0.0): The value used to pad the log-mel spectrograms to the same length in a batch. audio_token_duration_s (`float`, *optional*, defaults to 0.05): Duration, in seconds, represented by a single audio token, i.e. the STFT hop length. window_size_multiplier (`float`, *optional*, defaults to 2.0): Multiplier applied to `audio_token_duration_s` to obtain the STFT window length. n_fft (`int`, *optional*): FFT size. Defaults to the window length (`audio_token_duration_s * window_size_multiplier * sampling_rate`) when not provided. input_featuresinput_features_maskN feature_size sampling_rate padding_valueaudio_token_duration_swindow_size_multipliern_fftc zt |d|||d|||_||_t ||zd|_t ||z|zd|_|xs |j |_|j dks|j dks|jdkr tdtj|j dtj|_ t|jdzd z|d |d z |d d }tjtj |j"tj|_y)N)rrr z&audio_token_duration_s * sampling_ratez?audio_token_duration_s * window_size_multiplier * sampling_raterz7hop_length, window_size, and n_fft must all be positiveT)periodicdtype@slaney)num_frequency_binsnum_mel_filters min_frequency max_frequencyrnorm mel_scale)r&)super__init__r!r"r hop_length window_sizer#rr hann_windowfloat32windowr from_numpynpascontiguousarrayT mel_filters) selfrrr r!r"r#kwargsr> __class__s rr4z InklingFeatureExtractor.__init__Bs8  %''   '=#&<#' "] 24\ ) "%; ;m K M .d.. ??a 4#3#3q#8DJJ!OVW W''(8(84u}}] &#zzQ2('#-' !++B,@,@VXV`V`,abrwaveformdevicerc tj|jd|jz |jz|jdz }t |j |jz d}t j|||f}tj||j |j|j|jj|dd}tj|}|jdjdj!dj#}|j$j|}||z}|j!dj'}|j)ddS) NrFT)r5 win_lengthr9centerreturn_complexr'g|=r()mathceilshaper5maxr#Fpadrstftr6r9to view_as_realpowsum clamp_minsqrtr>log10 transpose) r?rBrC right_padleft_padrO magnitudesr>mel_specs r_torch_extract_fbank_featuresz5InklingFeatureExtractor._torch_extract_fbank_featuresos6IIhnnR04??BCdooUX`XfXfgiXjj tzzDOO3Q755Hi#89zz  JJ'';;>>&) ''- ^^A&**2.88?DDF &&))&1 +%%e,224!!!Q''r raw_speechpadding max_length truncationpad_to_multiple_ofreturn_attention_maskreturn_tensorsc b|8||jk7r>td|jd|jd|dtjd|jj ddd t jffd } t|tjrt j|}t|t jr,|jd kDrtd |jd |g} nt|ttfr`t|dk(r tdt|dt t"tj$tj&fr|g} n&t|} nt)ddt+|| D cgc]} | | dddf}} |Dcgc] }t|}}t-||d}|j/|||||d}|j0j3d}|j5|| }t j6|j8|j:zdz |j:d}t j<|j>d| dddf|dddfk}||jAdz}d|i}|r||d<t-||Scc} wcc}w)a Extract log-mel spectrogram features from one or several audio clip(s). Args: raw_speech (`np.ndarray`, `list[float]`, `list[np.ndarray]`, `list[list[float]]`): The sequence or batch of sequences to be padded. Each sequence can be a numpy array, a list of float values, a list of numpy arrays or a list of list of float values. Must be mono channel audio at `self.sampling_rate`, not stereo, i.e. single float per timestep. Decoding and resampling of raw audio (bytes / paths / URLs) is handled upstream by the processor's `apply_chat_template`, not here. sampling_rate (`int`, *optional*): The sampling rate of `raw_speech`, used only to validate against `self.sampling_rate`. device (`str`, *optional*, defaults to `"cpu"`): The device on which the log-mel spectrogram is computed in `_torch_extract_fbank_features`. NzWThe model corresponding to this feature extractor was trained using a sampling rate of zB. Please make sure that the provided audio input was sampled with z and not .zIt is strongly recommended to pass the `sampling_rate` argument to this function. Failing to do so can result in silent errors that might be hard to debug.clipz np.ndarray | torch.Tensor | listrct|tjr|n'tjt j |}|j tj}|jdk(r-tjdd|jd}|S|jdk7r"tdt|jd|S) Nr'z2Only mono-channel audio is supported for input to zB. Taking the mean over the channel (last) axis to convert to mono.rE)dimr(zDEach audio clip must be 1-D (mono) or 2-D (multichannel), got shape re) isinstancerTensor as_tensorr;asarrayrPr8ndimlogger warning_oncemeanrtuplerK)rftensorcls_names r_to_monoz2InklingFeatureExtractor.__call__.._to_monos'ell;TQSQ[Q[\`QaAbFYYu}}-F{{a##H SWW , M ! Z[`agamam[nZoopqMrr'zCA single array input must be 1-D (mono) or 2-D (multichannel); got z2 dims. Pass a list of arrays for a batch of clips.rzReceived an empty audio input.z!Unsupported audio input type for z: )r audio_lengthspt)r^r_r`rarcrEr(floor) rounding_mode)rCrr)data tensor_type)!rrrnrorA__name__rrjrir;ndarrayr:rmlistrqlenrfloatintegerfloating TypeErrortyperrNrsqueezer\divrur5arangerK unsqueeze)r?r]rr^r_r`rarbrcrCr@rtclipsrfspeechrubatched_speech padded_inputsinput_waveformsr num_framesrryrss @r__call__z InklingFeatureExtractor.__call__s8  $ 2 22 #1123((,(:(:';9]OSTV   \  >>** = %,,  j"** -))*5J j%,, /" YZdZiZiYjkBB LE  T5M 2:!# !ABB*Q-#ubjj"++)NO# Z(?zDQ[L\K]^_ _:?@$htnQW-@ @4>>V> >%Vc&de !!1 ! (66>>rB;;OVTYY  ' '$// 9A =t^e $ll>+?+?+B6RSWYZSZ[^hijlpip^qq'*=*G*G*KK .1 +>D& '>BB?A?s J'$J,)Pi>r)g?r*N)cpu)NTNFNTNr)r{ __module__ __qualname____doc__model_input_namesrrr4rrjstrr\r;r|r}boolrrrr __classcell__)rAs@rrr%s2*+@A#"(,(+ +c+c+c +c !& +c !& +cTz+cZ(ell(C(\a\h\h(6%)04!% )--126"nCJJe,tBJJ/??$tE{BSSnCTznCo- nC $J nC  nC $JnC $d{nCj(4/nCd nC nCrr)gư>)rInumpyr;rtorch.nn.functionalnn functionalrM audio_utilsr!feature_extraction_sequence_utilsrfeature_extraction_utilsrutilsrrr utils.import_utilsr get_loggerr{rnrrrrr__all__r2rrrs  *I499*   H %ces :OC6OCOCd % %r