o xi@s|ddlZddlZddlmZedjZddZddZ ej ddd d Z dd d Z ddZ ddZdddZddZdS)N) resample_polyfloatc Cst||}|dkr||}||}d}ddt||}|d}d|}t|dd|}t| |d}d||td||} |d kr_|d kr_d |d d d |d } n |d krjd|d} nd} td|d| | } | S)zPort of Octave code to Pythong? igX9ȶ<@2gW2ı?g?gUj@+0?gK46?gffffff!@g)npgcdmaxceilarangesinckaiser) pqr log10_rejectionstopband_cutoff_froll_off_width rejection_dBLt ideal_filterbetahr@/home/ubuntu/.local/lib/python3.10/site-packages/pystoi/utils.py_resample_window_oct s,   rcCs(t||}|t|}t||||dS)z(Resampler that is compatible with Octave)window)rr sumr)xrrrr rrr resample_oct.s r#)maxsizecCstd||d}|dt|dd}tt|t}td||}|tdd|dd}|tdd|dd}t|t |f} tt |D]2} t t ||| } || || <| } t t ||| } || || <| } d| | | | f<qR| |fS)aJ Returns the 1/3 octave band matrix and its center frequencies # Arguments : fs : sampling rate nfft : FFT size num_bands : number of 1/3 octave bands min_freq : center frequency of the lowest 1/3 octave band # Returns : obm : Octave Band Matrix cf : center frequencies rrNrgr(?g@) r linspaceintarrayrangeastyperpowerzeroslenargminsquare)fsnfft num_bandsmin_freqfkcffreq_low freq_highobmif_binfl_iifh_iirrrthirdoct5s    r>csTt|}tdddtfddtdt|D}|S)a Short-time Fourier transform for real 1-D inputs # Arguments x : 1D array, the waveform win_size : integer, the size of the window and the signal frames fft_size : integer, the size of the fft in samples (zero-padding or not) overlap: integer, number of steps to make in fftsize # Returns stft_out : 2D complex array, the STFT of x. rrcs,g|]}tjj||dqS))n)r fftrfft.0r:fft_sizewwin_sizer"rr bs$zstft..r)r'r hanningr(r)r-)r"rIrGoverlaphopstft_outrrFrstftVs rOcCs|j\}}| | }t|d|fd|||ff}|||||f}t|gd}|d|f}|d| }||||d|f}tj|dd}t|d||}|dd|}|S)Nr)rrrr@raxis)shaper padreshape transposer!r-)x_framesrM num_framesframelensegmentssignalendrrr_overlap_and_addgs   r\c stdddtfddtdt|D}tfddtdt|D}dttjj|dd t}t |||dk}||}||}t ||} t ||} | | fS) a9 Remove silent frames of x and y based on x A frame is excluded if its energy is lower than max(energy) - dyn_range The frame exclusion is based solely on x, the clean speech signal # Arguments : x : array, original speech wav file y : array, denoised speech wav file dyn_range : Energy range to determine which frame is silent framelen : Window size for energy evaluation hop : Hop size for energy evaluation # Returns : x without the silent frames y without the silent frames (aligned to x) rrr@c g|] }||qSrrrD)rXrHr"rrrJ z(remove_silent_frames..rcr]rrrD)rXrHyrrrJr^rP) r rKr(r)r-log10linalgnormEPSr r\) r"r_ dyn_rangerXrMrVy_frames x_energiesmaskx_sily_silr)rXrHr"r_rremove_silent_framess$$  rkr@cCstjt||ddS)zL Returns an array of vectors of norms of the rows of matrices from 3D array TrQkeepdims)r r!r/)r"rQrrr vect_two_normsrncs|ttj|j}|tj|ddd8}dtt|tfddt jdD}t ||}|t tj|j7}|tj|ddd8}dtt|dd tfd dt jdD}t ||}|S) zD Row and column mean and variance normalize an array of 2D segments r@Trlrc g|] }t|dqSr@r diagrTrDx_invrrrJr^z%row_col_normalize..rrrPcrorprqrDrsrrrJr^) rdr randomstandard_normalrRmeansqrtrnr(r)matmul)r"x_normedx_diagsrrsrrow_col_normalizes  r|)r?rp) functoolsnumpyr scipy.signalrfinfoepsrdrr# lru_cacher>rOr\rkrnr|rrrrs  %    '