o i @sjdZddlZddlZddlmmZddlm Z  dde de d e d ej fd d Z Gd ddejjZdS)zaPseudo QMF modules. This code is modified from https://github.com/kan-bayashi/ParallelWaveGAN. N)kaiser>V-?"@taps cutoff_ratiobetareturncCs|ddks Jdd|krdksJdJdtj|}tjdd&t|t|d d |tjt|d d |}Wd n1sPwYtd|||d<t|d |}||}|S) aKDesign prototype filter for PQMF. This method is based on `A Kaiser window approach for the design of prototype filters of cosine modulated filterbanks`_. Args: taps (int): The number of filter taps. cutoff_ratio (float): Cut-off frequency ratio. beta (float): Beta coefficient for kaiser window. Returns: ndarray: Impluse response of prototype filter (taps + 1,). .. _`A Kaiser window approach for the design of prototype filters of cosine modulated filterbanks`: https://ieeexplore.ieee.org/abstract/document/681427 rz'The number of taps mush be even number.?z%Cutoff ratio must be > 0.0 and < 1.0.ignore)invalidg?N)nppierrstatesinarangecosr)rrromega_ch_iwhrO/home/ubuntu/.local/lib/python3.10/site-packages/espnet2/gan_tts/melgan/pqmf.pydesign_prototype_filters$ rc sheZdZdZ    ddededed effd d Zd ejd ejfddZ d ejd ejfddZ Z S)PQMFzPQMF module. This module is based on `Near-perfect-reconstruction pseudo-QMF banks`_. .. _`Near-perfect-reconstruction pseudo-QMF banks`: https://ieeexplore.ieee.org/document/258122 rrrsubbandsrrrc stt|||}t|t|f}t|t|f}t|D]V}d|td|dtjd|t |d|dd|tjd||<d|td|dtjd|t |d|dd|tjd||<q!t | d} t | d} |d| |d| t |||f } t|D] }d| ||df<q|d | ||_t j|dd |_d S) aInitilize PQMF module. The cutoff_ratio and beta parameters are optimized for #subbands = 4. See dicussion in https://github.com/kan-bayashi/ParallelWaveGAN/issues/195. Args: subbands (int): The number of subbands. taps (int): The number of filter taps. cutoff_ratio (float): Cut-off frequency ratio. beta (float): Beta coefficient for kaiser window. r rrranalysis_filtersynthesis_filterr updown_filterr N)super__init__rrzeroslenrangerrrtorch from_numpyfloat unsqueezeregister_bufferrnn ConstantPad1dpad_fn) selfrrrrh_proto h_analysis h_synthesiskr!r"r# __class__rrr%AsV             z PQMF.__init__xr cCs(t|||j}tj||j|jdS)zAnalysis with PQMF. Args: x (Tensor): Input tensor (B, 1, T). Returns: Tensor: Output tensor (B, subbands, T // subbands). stride)Fconv1dr0r!r#rr1r8rrranalysiss z PQMF.analysiscCs.tj||j|j|jd}t|||jS)zSynthesis with PQMF. Args: x (Tensor): Input tensor (B, subbands, T // subbands). Returns: Tensor: Output tensor (B, 1, T). r9)r;conv_transpose1dr#rr<r0r"r=rrr synthesiss zPQMF.synthesis)rrrr) __name__ __module__ __qualname____doc__intr+r%r)Tensorr>r@ __classcell__rrr6rr7s" A r)rrr)rDnumpyrr)torch.nn.functionalr. functionalr; scipy.signalrrEr+ndarrayrModulerrrrrs   '