AVhaNbdZddlZddlmZddlmZddlmZddlmZddlm Z ddlm Z dd l m Z dd l m Z dd l mZdd l mZdd l mZddlmZddlmZedej*dej,ddfdZedej*ej,dfdZedej*dej,dfdZdZedej*ej6dddfdZedej*ej6ddfdZy)z8Spectral operations (e.g. Short-time Fourier Transform).N) constant_op)dtypes)ops) tensor_util) array_ops)math_ops)dct_ops)fft_ops)reconstruction_ops) shape_ops) window_ops)dispatch) tf_exportz signal.stftFcntj|d|||g5tj|d}|jj dtj|d}|jj dtj|d}|jj d| t |}ntj|d }tj|||| }||||j }||z}tj||gcdddS#1swYyxYw) aCComputes the [Short-time Fourier Transform][stft] of `signals`. Implemented with TPU/GPU-compatible ops and supports gradients. Args: signals: A `[..., samples]` `float32`/`float64` `Tensor` of real-valued signals. frame_length: An integer scalar `Tensor`. The window length in samples. frame_step: An integer scalar `Tensor`. The number of samples to step. fft_length: An integer scalar `Tensor`. The size of the FFT to apply. If not provided, uses the smallest power of 2 enclosing `frame_length`. window_fn: A callable that takes a window length and a `dtype` keyword argument and returns a `[window_length]` `Tensor` of samples in the provided datatype. If set to `None`, no windowing is used. pad_end: Whether to pad the end of `signals` with zeros when the provided frame length and step produces a frame that lies partially past its end. name: An optional name for the operation. Returns: A `[..., frames, fft_unique_bins]` `Tensor` of `complex64`/`complex128` STFT values where `fft_unique_bins` is `fft_length // 2 + 1` (the unique components of the FFT). Raises: ValueError: If `signals` is not at least rank 1, `frame_length` is not scalar, or `frame_step` is not scalar. [stft]: https://en.wikipedia.org/wiki/Short-time_Fourier_transform stftsignalsname frame_lengthr frame_stepN fft_lengthpad_enddtype) r name_scopeconvert_to_tensorshapewith_rank_at_leastassert_has_rank_enclosing_power_of_twor framerr rfft) rrrr window_fnrrframed_signalswindows Y/home/dcms/DCMS/lib/python3.12/site-packages/tensorflow/python/ops/signal/spectral_ops.pyrr"sD ~~dFWl%/%126##G).inverse_stft_window_fn_innerws0 69J8K L3))*<Hk''***(M$N#OPe+'>?e!!%T:ennUXqM2e;(>'?@e eM\2 2!333s DD77E)rr6rr7s``` r(r+r+_s0 3B &%zsignal.inverse_stftc ^tj|d|g5tj|d}|jj dtj|d}|jj dtj|d}|jj d| t |}n2tj|d }|jj dtj||g}tj|}|3|jj|jjd |d d|f}tj|}tj|} tjtj |d z dg|j" dt%j&d|| d z gggd} tj(|| }n|jjd |kDr |d d|f}ny|jjd |krY||jjd z } tj(|ddgg|jjd z zd| ggz}|F|jj0|j+dg|jjd z z|gz|#|||j"j, } || z}t/j0||cdddS#1swYyxYw)aF Computes the inverse [Short-time Fourier Transform][stft] of `stfts`. To reconstruct an original waveform, a complementary window function should be used with `inverse_stft`. Such a window function can be constructed with `tf.signal.inverse_stft_window_fn`. Example: ```python frame_length = 400 frame_step = 160 waveform = tf.random.normal(dtype=tf.float32, shape=[1000]) stft = tf.signal.stft(waveform, frame_length, frame_step) inverse_stft = tf.signal.inverse_stft( stft, frame_length, frame_step, window_fn=tf.signal.inverse_stft_window_fn(frame_step)) ``` If a custom `window_fn` is used with `tf.signal.stft`, it must be passed to `tf.signal.inverse_stft_window_fn`: ```python frame_length = 400 frame_step = 160 window_fn = tf.signal.hamming_window waveform = tf.random.normal(dtype=tf.float32, shape=[1000]) stft = tf.signal.stft( waveform, frame_length, frame_step, window_fn=window_fn) inverse_stft = tf.signal.inverse_stft( stft, frame_length, frame_step, window_fn=tf.signal.inverse_stft_window_fn( frame_step, forward_window_fn=window_fn)) ``` Implemented with TPU/GPU-compatible ops and supports gradients. Args: stfts: A `complex64`/`complex128` `[..., frames, fft_unique_bins]` `Tensor` of STFT bins representing a batch of `fft_length`-point STFTs where `fft_unique_bins` is `fft_length // 2 + 1` frame_length: An integer scalar `Tensor`. The window length in samples. frame_step: An integer scalar `Tensor`. The number of samples to step. fft_length: An integer scalar `Tensor`. The size of the FFT that produced `stfts`. If not provided, uses the smallest power of 2 enclosing `frame_length`. window_fn: A callable that takes a window length and a `dtype` keyword argument and returns a `[window_length]` `Tensor` of samples in the provided datatype. If set to `None`, no windowing is used. name: An optional name for the operation. Returns: A `[..., samples]` `Tensor` of `float32`/`float64` signals representing the inverse STFT for each input STFT in `stfts`. Raises: ValueError: If `stfts` is not at least rank 2, `frame_length` is not scalar, `frame_step` is not scalar, or `fft_length` is not scalar. [stft]: https://en.wikipedia.org/wiki/Short-time_Fourier_transform inverse_stftstftsrrrrNr.rr)rrrrr r!r"r irfftrconstant_valuendimsas_listrrankconcatzerosrrmaximumr. set_shape real_dtyper overlap_and_add) r<rrrr%r real_framesframe_length_staticreal_frames_rankreal_frames_shapepaddings pad_amountr's r(r;r;sF ~~dNUG44G  ! !%g 6E KK""1%((NKL&&q)&&z EJ$$Q'*<8j((,Gj&&q)-- |4K&44\B #{'8'8'>'>'F!!#B'/]l] 23k" 4#//+6!! ??,q0!4!-!3!3 5   L3DR3H$HI J K MNOQhMM+x8k    " " $R (+> >%9&9%9 9:k    " " $R (+> >&):):)B)B)DR)HHjMM+$%q6(k.?.?.E.E.I"J$%z?"3#45k&;+<+<+B+B+NTFk&7&7&=&=&AB0123 U[[-C-CDfVk  - -k: Fi4G4G4Gs LL##L,c $tj|}|htjt dt j t j|t jdz z|jStjtjdtj tjtj|tjtjdz |jS)z2Return 2**N for integer N such that 2**N >= value.r=g@)rr@rconstantintnpceillogrrcastpowrfloat32)value value_statics r(r"r"s++E2,    Arwwrvvl+bffSk9: :;U[[ JJ ll --ll8==?@ll3  !"$);;  00r9z signal.mdctctj|d||g5tj|d}|jj dtj|d}|jj dt j|}|8|dzdk7r td tj|d z|j }n|d z}tj|||| }||||j } || z}n|d tjd z z}tj|dd} tj | d dg | dz } | dtj | ddgz } tj"| | fd} t%j&| d|cdddS#1swYyxYw)aComputes the [Modified Discrete Cosine Transform][mdct] of `signals`. Implemented with TPU/GPU-compatible ops and supports gradients. Args: signals: A `[..., samples]` `float32`/`float64` `Tensor` of real-valued signals. frame_length: An integer scalar `Tensor`. The window length in samples which must be divisible by 4. window_fn: A callable that takes a frame_length and a `dtype` keyword argument and returns a `[frame_length]` `Tensor` of samples in the provided datatype. If set to `None`, a rectangular window with a scale of 1/sqrt(2) is used. For perfect reconstruction of a signal from `mdct` followed by `inverse_mdct`, please use `tf.signal.vorbis_window`, `tf.signal.kaiser_bessel_derived_window` or `None`. If using another window function, make sure that w[n]^2 + w[n + frame_length // 2]^2 = 1 and w[n] = w[frame_length - n - 1] for n = 0,...,frame_length // 2 - 1 to achieve perfect reconstruction. pad_end: Whether to pad the end of `signals` with zeros when the provided frame length and step produces a frame that lies partially past its end. norm: If it is None, unnormalized dct4 is used, if it is "ortho" orthonormal dct4 is used. name: An optional name for the operation. Returns: A `[..., frames, frame_length // 2]` `Tensor` of `float32`/`float64` MDCT values where `frames` is roughly `samples // (frame_length // 2)` when `pad_end=False`. Raises: ValueError: If `signals` is not at least rank 1, `frame_length` is not scalar, or `frame_length` is not a multiple of `4`. [mdct]: https://en.wikipedia.org/wiki/Modified_discrete_cosine_transform mdctrrrrrNz)The frame length must be a multiple of 4.r=rr?r>axistypenorm)rrrrr r!rr@ ValueErrorrr r#rSsqrtrsplitreverserDr dct)rrr%rrdrrKrr&r' split_framesframe_firsthalfframe_secondhalfframes_rearrangeds r(r\r\#sN ~~dFWl$;<"=##G)12>L ((a*,//1=aAO#A):):<?<>4*AA!((/;K)L.02 ;;(qt >> @tf.function ... def compare_round_trip(): ... samples = 1000 ... frame_length = 400 ... halflen = frame_length // 2 ... waveform = tf.random.normal(dtype=tf.float32, shape=[samples]) ... waveform_pad = tf.pad(waveform, [[halflen, 0],]) ... mdct = tf.signal.mdct(waveform_pad, frame_length, pad_end=True, ... window_fn=tf.signal.vorbis_window) ... inverse_mdct = tf.signal.inverse_mdct(mdct, ... window_fn=tf.signal.vorbis_window) ... inverse_mdct = inverse_mdct[halflen: halflen + samples] ... return waveform, inverse_mdct >>> waveform, inverse_mdct = compare_round_trip() >>> np.allclose(waveform.numpy(), inverse_mdct.numpy(), rtol=1e-3, atol=1e-4) True Implemented with TPU/GPU-compatible ops and supports gradients. Args: mdcts: A `float32`/`float64` `[..., frames, frame_length // 2]` `Tensor` of MDCT bins representing a batch of `frame_length // 2`-point MDCTs. window_fn: A callable that takes a frame_length and a `dtype` keyword argument and returns a `[frame_length]` `Tensor` of samples in the provided datatype. If set to `None`, a rectangular window with a scale of 1/sqrt(2) is used. For perfect reconstruction of a signal from `mdct` followed by `inverse_mdct`, please use `tf.signal.vorbis_window`, `tf.signal.kaiser_bessel_derived_window` or `None`. If using another window function, make sure that w[n]^2 + w[n + frame_length // 2]^2 = 1 and w[n] = w[frame_length - n - 1] for n = 0,...,frame_length // 2 - 1 to achieve perfect reconstruction. norm: If "ortho", orthonormal inverse DCT4 is performed, if it is None, a regular dct4 followed by scaling of `1/frame_length` is performed. name: An optional name for the operation. Returns: A `[..., samples]` `Tensor` of `float32`/`float64` signals representing the inverse MDCT for each input MDCT in `mdcts` where `samples` is `(frames - 1) * (frame_length // 2) + frame_length`. Raises: ValueError: If `mdcts` is not at least rank 2. [mdct]: https://en.wikipedia.org/wiki/Modified_discrete_cosine_transform inverse_mdctmdctsrr=r>rNg?r])rcorthorbr_rrr^)rrrrr rrVrint32rr rirrgrDrhrSrfr rI) rpr%rdrhalf_lenhalf_len_float result_idct4 split_resultrJr's r(roroosxv ~~dNUG4E  ! !%g 6E KK""1%}}U[[_FLLAH |}}XU[[AnN*gkk%a.HHl [[QW=l??<F!!F*)__doc__numpyrStensorflow.python.frameworkrrrrtensorflow.python.opsrrtensorflow.python.ops.signalr r r r r tensorflow.python.utilr tensorflow.python.util.tf_exportradd_dispatch_support hann_windowrr+r;r" vorbis_windowr\ror8r9r(rsZ?3.+3+*00;23+6 = 7;))T8686v *+ -7-C-C $7&,7&t ! !%11 uG"uGp 0 = *4*B*BTG=G=T ! %33PE"PEr9