iSbdZddlZgdZddlmZddlmZddlmZ edd6dZ edd6dZ edd6d Z edd6d Z e e e e e d Z d7d e jezfd Z d8dZed d9dededededeedeedede jfdZGddZ d:de jdedede jfdZde jde jfdZd e jde jfd!Zd"edefd#Zd$e jd%ed&ed'ede jf d(Zd)ed*ed+ed,ed-edee je jff d.Z d;d$e jded/ed0ed1ed2ed3ed4ed'ed,ed-ede jfd5Zy)._padvs z !66!w012 2  "q7Q;'"-F1~+DbD1F>>61f"56 60 ;< <rrzInput is too short (length=z ) for n_fft=z with hop_length=z and center=.shapestrides)r5) isinstancestrrgetlowerr9rCrr8zeros as_stridedfftrfft)r:n_fft hop_length win_lengthr.centerr< window_fnwpad_sizer? num_framesrCrDframess r$rr\s_aZ  &#$((8  8AB B j ! wwqzE1771:% NNArxx 45A >= EQJ )aggaj5(Z77JQ)!''!*\%I$\& <   E1oG ]]1E7 ;F 66;;vz ""r@cb||jddz dz}||dz}t|trBtj |j }|t d|||dzdd}n|}|jd|kr=tj|tj||jdz fgd}|jd} | dz |z|z} tj| } tj| } tjj|djdd} tj| |z}|dddftj|z}|j}| |zj}|r||zn|}tj|| fj}| j |j#|} | j |j#|} tj$| d kD| | z | } |r| | |dz| dz} || d|} | S) akInverse Short-Time Fourier Transform. Args: x: Complex STFT output of shape (n_fft // 2 + 1, num_frames) hop_length: Hop length between frames (default: win_length // 4) win_length: Window length (default: (n_fft - 1) * 2) window: Window function name or array (default: "hann") center: If True, remove center padding (default: True) length: Target output length (default: None) normalized: If True, use window squared (COLA) normalization. If False, use simple window normalization (default: True) Returns: Reconstructed time-domain signal Nrrr*r0r6rr1绽|=)rCrErFrrGrHr9rr8rIrKirfft transposearangeflattentileataddwhere)r:rNrOr.rPlength normalizedrQrRrTt reconstructed window_sum frames_time frame_offsetsindices indices_flatupdates_reconstructed window_normupdates_windows r$r r s'.ggaj1n) 1_ &#$((8  8AB B j1n %cr * wwqzJ NNArxxaggaj)@(BCD1 MJ a:% 2AHHQKM!J&&,,qq,)33Aq9KIIj)J6MAtG$ryy'<.hz_to_mels  DJJsTE\'9:: :% tu $ !E)T1 ((3-$& : $*;!.mel_to_hzs  DTF]3c9: :% tt # !E)T1 ((3-$& K  4++= >? ?    r@rrrr6slaney@)ru)rlinspace expand_dimsmaximum zeros_likeminimummoveaxis)rlrMrmrnrorprqrrn_freqs all_freqsm_minm_maxm_ptsf_ptsf_diffslopes down_slopes up_slopes filterbankenorms r$r r sn "  $[1_Eqj1nG A{a/9I eY 'E eY 'E KKufqj 1E eY 'E12Ys #F ^^E1 %y!(D DF1crc6N?fSbk1Kq!"u qr *I k"BJJ{I$FJ xuQ!,uWf~=>bnnUA.. $$Q*J r@ceZdZdZdZdededefdZdededed ejdef d Z dd ejd ejdededed ejde dedejfdZ dZ dZ y )r z Advanced caching for iSTFT operations. Fully vectorized Overlap-Add for MLX. Handles multiple configurations efficiently. Automatically caches normalization buffers and position indices for maximum performance. c i|_i|_y)N)norm_buffer_cacheposition_cacheselfs r$__init__zISTFTCache.__init__2s!# r@rT frame_lengthrNc|||f}||jvrZtj|dddf|ztj|dddfz}|jd|j|<|j|S)z.Get cached position indices or create new onesNr6)rrrZreshape)rrTrrNkey positionss r$ get_positionszISTFTCache.get_positions6s<4 d)) ) *%ag.;))L)$'23 (1'8'8'1 ffll<#9#9!Q#BeRTlU &./>/D/D, J >Z/,>** :z6: ++J jQ *-7 'a0=D3II:/ C>11"56::?;R;RSU;VW G4+dAg..  IAyzM*F  #A} },-F r@cl|jj|jjy)z$Clear all cached data to free memoryN)rclearrrs r$ clear_cachezISTFTCache.clear_caches& $$& !!#r@ct|jt|jt|jt|jzdS)z"Get information about cached items) norm_buffersposition_indicestotal_cached_items)lenrrrs r$ cache_infozISTFTCache.cache_infosK  6 67 #D$7$7 8"%d&<&<"=$%%&#'  r@)TN)__name__ __module__ __qualname____doc__rintrrrrboolr rrr@r$r r +s ! ( (3 (C (+++ +  +  +F C88C88C C  C  CCCC CJ$  r@r specgramrOmodec0|dkrtd||j}|jd|d}|jd}|dz dz}t||dzzd|zdzzdz }|dk(r]t j |d d ddf|d }t j |d d dd f|d }t j |||gd } nt j|d ||fg} t j| |dz| j } | jdd|zz } t j|| f| j } t| D]6} | d d | | |zf}|| z}t j|d |z | d d | f<8| j|S) a Compute delta coefficients of a spectrogram (Kaldi-compatible). The formula is: d_t = sum_{n=1}^{N} n * (c_{t+n} - c_{t-n}) / (2 * sum_{n=1}^{N} n^2) Args: specgram: MLX array of dimension (..., freq, time) win_length: The window length used for computing delta (default: 5) mode: Padding mode - "edge" or "constant" (default: "edge") Returns: MLX array of deltas of dimension (..., freq, time) zwin_length should be >= 3, got r6rrrg@edgeNr1rrr) r9rCrfloatrrepeatr8r7rZrrIrsum)rrOroriginal_shape num_featuresr#r"pad_left pad_rightpaddedkernel_weights time_stepsrir.weighteds r$r r s"A~::,GHH^^NN2$67H>>!$L aAA !q1u+Q+ ,s 2E v~99Xa1f-qq9IIhq"#v.: 8Y ?aH6Aq6"23YYr1q5 =Na1q5(J XX|Z0 EF : 1q:~--.N*vvhQ/%7q!t  >>. ))r@rc>dtjd|dz zzS)z/Convert frequency to mel scale (Kaldi formula).@rwrx)rr)rs r$r r s BFF3-. ..r@mel_freqc>dtj|dz dz zS)z/Convert mel scale to frequency (Kaldi formula).rxrrw)rr)rs r$rrs BFF8f,-3 44r@r:c<|dk(rdSd|dz jzS)z7Returns the smallest power of 2 that is greater than x.rrr) bit_length)r:s r$_next_power_of_2rs%Q15A!a%!3!3!555r@waveform window_size window_shift snip_edgesc|jd}|r&||krtjdSd||z |zz}n~||dzz|z}|dz|dzz }|dkDr@|d|dzddd}|dkDr |d| dz dn|ddd}tj|||g}n#|ddd}tj|| d|g}tj|||f|dfS)zCExtract frames from waveform using strided windowing (Kaldi-style).rrrrNr6rB)rCrrIr8rJ) rrrr num_samplesmr7rrs r$_get_strided_kaldirs..#K  $88F# # {*|; ; LA- .< ?Q!22 7C!G,TrT2H8;asdQh!34XbQRSUgEVI~~x9&EFH 2I~~x &BCH ==![)9LRSCT UUr@num_binswindow_length_padded sample_freqlow_freq high_freqc|dkDsJd|dzdk(sJ|dz}d|z}|dkr||z }d|cxkr|krnJd|cxkr|ksJJ||z }tttj|}tttj|} | |z |dzz } tj|j dd} || | zz} || d z| zz} || d z| zz}t | }t|tj|zj dd}|| z | | z z }||z || z z }tjtjdtj||}||jfS) a Create Kaldi-compatible mel filterbank matrix. Args: num_bins: Number of mel bins window_length_padded: Padded window length (FFT size) sample_freq: Sample frequency in Hz low_freq: Low frequency cutoff high_freq: High frequency cutoff (0 or negative = relative to Nyquist) Returns: (bins, center_freqs): Mel filterbank matrix and center frequencies rzMust have at least 3 mel binsrrrrzrr6rwr) rr rrrZrrrrIrsqueeze)rrrrr num_fft_binsnyquist fft_bin_width mel_low_freq mel_high_freqmel_freq_deltabin_idxleft_mel center_mel right_mel center_freqsmelup_slope down_slopebinss r$rrs( a<888< ! #q (( ('1,LKGCW 8 %g %G GC),Fw,FG G,FG G"66M();<=L/"((9*=>?M#l2x!|DNii!))"a0Gg66H3. @@J# ??I*:6L -"))L*AA B J J1b QCh:#89Hc/i*&<=J ::bhhqk2::h #C DD %%' ''r@win_lenwin_incnum_melswin_type preemphasisditherc 6|jdk(r|d}||z dz} ||z dz} t|| zdz} t|| zdz}t|}t||| |}|jddk(rt j d|fS|dk7r1t jj|j|z}||z}t j|dd}||z }|dk7r>|d d ddf}|d d dd f||d d d d fzz }t j||gd }|d k(rKt j|}d dt jdt jz|z|dz z zz }n|dk(rKt j|}ddt jdt jz|z|dz z zz }n{|dk(rat j|}ddt jdt jz|z|dz z zz }t j|d}nt j|}||z}||k7r||z }t j |dd|fg}t j"j%||d}t j&|dz}t)||t+|| | \}}t j |ddg}t j,||j.}t j0t j2|d}|S)a Compute Kaldi-compatible log mel-filterbank features. Args: waveform: Input audio (1D array) sample_rate: Sample rate in Hz win_len: Window length in samples win_inc: Window shift in samples num_mels: Number of mel bins win_type: Window type ("hamming", "hanning", "povey", "rectangular") preemphasis: Preemphasis coefficient dither: Dither amount (0 to disable) snip_edges: If True, discard incomplete frames at edges low_freq: Low frequency cutoff high_freq: High frequency cutoff (0 = Nyquist) Returns: Log mel-filterbank features (time, num_mels) rrigMbP?rzrT)r2keepdimsNr6r1rr'r(rrpoveyg333333?rrr)rrg:0yE>)ndimrrrrCrrIrandomnormalmeanr8rZrrpoweronesr7rKrLr,rrmatmulTrr)rrlrrrrrrrrrframe_length_msframe_shift_mswindow_shift_samplesrpadded_window_size strided_input rand_gauss row_means first_col other_colsr#r.r-r; fft_resultspectrum mel_energies_ mel_featuress r$rrAs @}}A; +d2O{*T1N{^;eCDkO3e;QG 9 IIk "rvva"%%i!m{Q&GHHH Y  IIk "sRVVAIM[1_$EFFF W  IIk "S266!bee)a-;?"CDDD$%%!F*M[(${2}v7|.DE ].@qIJvvj!S(H*$eK&8(IOL!66,(89L99X|~~6L66"**\489L r@)F)i NNr-Tr5)NNr-TNF)rNNru)r) iii<rg ףp= ?rwTg4@rz) rr__all__ functoolsrtypingrmlx.corecorerrrrrrrrFrr rrr r r r rrrrrrrrr@r$r5s4   "  4   4   4   4OO   #  1# HHsN 1#l   I^ 4 !HH H H  H E? H 3- HHXXHHVD D Z:@,*hh,*$',*36,*XX,*^/"((/rxx/ 5bhh52885 666 VhhV%(V8;VIMVXXV20(0(0(0( 0(  0(  288RXX  0(jchhccc c  c  cc ccccXXcr@