o jP@s2ddlZddlZddlmZGdddejZdS)N)nncsheZdZdZ             dfd d Zd d Zd dZedddZedddZ Z S) TorchSTFTa_Some of the audio processing funtions using Torch for faster batch processing. Args: n_fft (int): FFT window size for STFT. hop_length (int): number of frames between STFT columns. win_length (int, optional): STFT window length. pad_wav (bool, optional): If True pad the audio with (n_fft - hop_length) / 2). Defaults to False. window (str, optional): The name of a function to create a window tensor that is applied/multiplied to each frame/window. Defaults to "hann_window" sample_rate (int, optional): target audio sampling rate. Defaults to None. mel_fmin (int, optional): minimum filter frequency for computing melspectrograms. Defaults to None. mel_fmax (int, optional): maximum filter frequency for computing melspectrograms. Defaults to None. n_mels (int, optional): number of melspectrogram dimensions. Defaults to None. use_mel (bool, optional): If True compute the melspectrograms otherwise. Defaults to False. do_amp_to_db_linear (bool, optional): enable/disable amplitude to dB conversion of linear spectrograms. Defaults to False. spec_gain (float, optional): gain applied when converting amplitude to DB. Defaults to 1.0. power (float, optional): Exponent for the magnitude spectrogram, e.g., 1 for energy, 2 for power, etc. Defaults to None. use_htk (bool, optional): Use HTK formula in mel filter instead of Slaney. mel_norm (None, 'slaney', or number, optional): If 'slaney', divide the triangular mel weights by the width of the mel band (area normalization). If numeric, use `librosa.util.normalize` to normalize each filter by to unit l_p norm. See `librosa.util.normalize` for a full description of supported norm values (including `+-np.inf`). Otherwise, leave all the triangles aiming for a peak value of 1.0. Defaults to "slaney". F hann_windowNrP?slaneycst||_||_||_||_||_||_||_| |_ | |_ | |_ | |_ | |_ ||_||_tjtt||dd|_d|_||_| rJ|dSdS)NF) requires_grad)super__init__n_fft hop_length win_lengthpad_wav sample_ratemel_fminmel_fmaxn_melsuse_mel do_amp_to_db spec_gainpoweruse_htkmel_normr Parametergetattrtorchwindow mel_basis normalized_build_mel_basis)selfr r r rrrrrrrrrrrrr __class__S/home/kuhnn/.local/lib/python3.10/site-packages/TTS/utils/audio/torch_transforms.pyr @s*  zTorchSTFT.__init__c Cs|jdkr |d}|jr#t|j|jd}tjjj |||fdd}tj | d|j|j|j |j dd|jddd }|ddddddd f}|dddddddf}ttj|d|dd d }|jdurp||j}|jr}t|j||}|jr|j||jd }|S) zCompute spectrogram frames by torch based stft. Args: x (Tensor): input waveform Returns: Tensor: spectrogram frames. Shapes: x: [B x T] or [:math:`[B, 1, T]`] reflect)modeTF)centerpad_moderonesidedreturn_complexNrg:0yE>min)r)ndim unsqueezerintr r rr functionalpadstftsqueezer rrsqrtclamprrmatmulrtor _amp_to_dbr)r xpaddingoMPSr#r#r$__call__hs6     zTorchSTFT.__call__c Cs<tjj|j|j|j|j|j|j|j d}t | |_ dS)N)srr rfminfmaxhtknorm)librosafiltersmelrr rrrrrr from_numpyfloatr)r rr#r#r$rs zTorchSTFT._build_mel_basiscCsttj|dd|S)Ngh㈵>r-)rlogr7r;rr#r#r$r:szTorchSTFT._amp_to_dbcCst||S)N)rexprMr#r#r$ _db_to_ampszTorchSTFT._db_to_amp) FrNrNrFFrNFrF)r) __name__ __module__ __qualname____doc__r rAr staticmethodr:rO __classcell__r#r#r!r$rs,>(+ r)rGrrModulerr#r#r#r$s