͑i^SSKJr SSKJrJr SSKrSSKJr SSKJrJ r J r SSK J r SSK Jr SS KJrJrJr SS KJr \(a SS KJr \S rS S/rSSSjjrSSSjjrSSSjjrSSSjjrg)) annotations) TYPE_CHECKINGLiteralN)_C_ops)in_dynamic_modein_dynamic_or_pir_mode in_pir_mode)check_variable_and_dtype) LayerHelper)fft_c2cfft_c2rfft_r2c) is_complex)TensorrstftistftcUS;a[SUS35e[U[5(aUS::a[SUS35e[U[5(aUS::a[SUS35e[5(aHXRU:a[SUS URUS 35e[ R "XX#5$[5(a[ R "XX#5$S n[US /S QU5 [U40[5D6nURS S9nURUS9nURUSU0UUUS.SU0S9 U$)a Slice the N-dimensional (where N >= 1) input into (overlapping) frames. Args: x (Tensor): The input data which is a N-dimensional (where N >= 1) Tensor with shape `[..., seq_length]` or `[seq_length, ...]`. frame_length (int): Length of the frame and `0 < frame_length <= x.shape[axis]`. hop_length (int): Number of steps to advance between adjacent frames and `0 < hop_length`. axis (int, optional): Specify the axis to operate on the input Tensors. Its value should be 0(the first dimension) or -1(the last dimension). If not specified, the last axis is used by default. name (str|None, optional): The default value is None. Normally there is no need for user to set this property. For more information, please refer to :ref:`api_guide_Name`. Returns: The output frames tensor with shape `[..., frame_length, num_frames]` if `axis==-1`, otherwise `[num_frames, frame_length, ...]` where `num_frames = 1 + (x.shape[axis] - frame_length) // hop_length` Examples: .. code-block:: pycon >>> import paddle >>> from paddle import signal >>> # 1D >>> x = paddle.arange(8) >>> y0 = signal.frame(x, frame_length=4, hop_length=2, axis=-1) >>> print(y0) Tensor(shape=[4, 3], dtype=int64, place=Place(cpu), stop_gradient=True, [[0, 2, 4], [1, 3, 5], [2, 4, 6], [3, 5, 7]]) >>> y1 = signal.frame(x, frame_length=4, hop_length=2, axis=0) >>> print(y1) Tensor(shape=[3, 4], dtype=int64, place=Place(cpu), stop_gradient=True, [[0, 1, 2, 3], [2, 3, 4, 5], [4, 5, 6, 7]]) >>> # 2D >>> x0 = paddle.arange(16).reshape([2, 8]) >>> y0 = signal.frame(x0, frame_length=4, hop_length=2, axis=-1) >>> print(y0) Tensor(shape=[2, 4, 3], dtype=int64, place=Place(cpu), stop_gradient=True, [[[0 , 2 , 4 ], [1 , 3 , 5 ], [2 , 4 , 6 ], [3 , 5 , 7 ]], [[8 , 10, 12], [9 , 11, 13], [10, 12, 14], [11, 13, 15]]]) >>> x1 = paddle.arange(16).reshape([8, 2]) >>> y1 = signal.frame(x1, frame_length=4, hop_length=2, axis=0) >>> print(y1.shape) paddle.Size([3, 4, 2]) >>> # > 2D >>> x0 = paddle.arange(32).reshape([2, 2, 8]) >>> y0 = signal.frame(x0, frame_length=4, hop_length=2, axis=-1) >>> print(y0.shape) paddle.Size([2, 2, 4, 3]) >>> x1 = paddle.arange(32).reshape([8, 2, 2]) >>> y1 = signal.frame(x1, frame_length=4, hop_length=2, axis=0) >>> print(y1.shape) paddle.Size([3, 4, 2, 2]) rUnexpected axis: . It should be 0 or -1.rzUnexpected frame_length: #. It should be an positive integer.Unexpected hop_length: zKAttribute frame_length should be less equal than sequence length, but got (z) > (z).framex)int32int64float16float32float64input_param_namedtypeX) frame_length hop_lengthaxisOuttypeinputsattrsoutputs) ValueError isinstanceintrshaperrr r r locals input_dtype"create_variable_for_type_inference append_op) rr'r(r)nameop_typehelperr%outs M/var/www/html/banglarbhumi/venv/lib/python3.13/site-packages/paddle/signal.pyrr*s|d 7,TF2IJKK lC ( (LA,=' ~5X Y   j# & &*/%j\1T U   ''$- '(>qwwt}oRA ||AZ>> ||AZ>> sG W11""C"877e7D8 ,( CL   JcUS;a[SUS35e[U[5(aUS::a[SUS35eSn[5(a[R "XU5nU$[ US/S QU5 [U40[5D6nURSS 9nURUS 9nURUS U0XS .SU0S9 U$)a Reconstructs a tensor consisted of overlap added sequences from input frames. Args: x (Tensor): The input data which is a N-dimensional (where N >= 2) Tensor with shape `[..., frame_length, num_frames]` or `[num_frames, frame_length ...]`. hop_length (int): Number of steps to advance between adjacent frames and `0 < hop_length <= frame_length`. axis (int, optional): Specify the axis to operate on the input Tensors. Its value should be 0(the first dimension) or -1(the last dimension). If not specified, the last axis is used by default. name (str|None, optional): The default value is None. Normally there is no need for user to set this property. For more information, please refer to :ref:`api_guide_Name`. Returns: The output frames tensor with shape `[..., seq_length]` if `axis==-1`, otherwise `[seq_length, ...]` where `seq_length = (n_frames - 1) * hop_length + frame_length` Examples: .. code-block:: pycon >>> import paddle >>> from paddle.signal import overlap_add >>> # 2D >>> x0 = paddle.arange(16).reshape([8, 2]) >>> print(x0) Tensor(shape=[8, 2], dtype=int64, place=Place(cpu), stop_gradient=True, [[0 , 1 ], [2 , 3 ], [4 , 5 ], [6 , 7 ], [8 , 9 ], [10, 11], [12, 13], [14, 15]]) >>> y0 = overlap_add(x0, hop_length=2, axis=-1) >>> print(y0) Tensor(shape=[10], dtype=int64, place=Place(cpu), stop_gradient=True, [0 , 2 , 5 , 9 , 13, 17, 21, 25, 13, 15]) >>> x1 = paddle.arange(16).reshape([2, 8]) >>> print(x1) Tensor(shape=[2, 8], dtype=int64, place=Place(cpu), stop_gradient=True, [[0 , 1 , 2 , 3 , 4 , 5 , 6 , 7 ], [8 , 9 , 10, 11, 12, 13, 14, 15]]) >>> y1 = overlap_add(x1, hop_length=2, axis=0) >>> print(y1) Tensor(shape=[10], dtype=int64, place=Place(cpu), stop_gradient=True, [0 , 1 , 10, 12, 14, 16, 18, 20, 14, 15]) >>> # > 2D >>> x0 = paddle.arange(32).reshape([2, 1, 8, 2]) >>> y0 = overlap_add(x0, hop_length=2, axis=-1) >>> print(y0.shape) paddle.Size([2, 1, 10]) >>> x1 = paddle.arange(32).reshape([2, 8, 1, 2]) >>> y1 = overlap_add(x1, hop_length=2, axis=0) >>> print(y1.shape) paddle.Size([10, 1, 2]) rrrrrr overlap_addr)rrrr r!uint16r"r$r&)r(r)r*r+) r0r1r2rrr?r r r4r5r6r7)rr(r)r8r9r;r:r%s r<r?r?sT 7,TF2IJKK j# & &*/%j\1T U  G  5" J !  I   W11""C"877e7D8!+:CL  Jr=c ([UR5n U S;d SU 35eU S:XaURS5nUc[US-5nUS:d SUS35eUcUn[ 5(a7SUs=:aURS ::dO S URS S US35eSUs=:aU::dO S US US35eUbA[UR5S:Xa[U5U:XdS USURS35eO [ R "U4URS9nX1:a8X- S-n X- U - n [ RRRXKU /SS9nU(a_US;d SUS35eUS-n [ RRRURS 5X/USS9RS 5n[XUS S9nUR/SQS9n[ R"X5nU(aSOSnUc[U5(+n[U5(aU(aS5e[U5(d[!USS USUU S9nO[#USS USU S 9nUR/SQS9nU S:XaUR%S5 U$)!a Short-time Fourier transform (STFT). The STFT computes the discrete Fourier transforms (DFT) of short overlapping windows of the input using this formula: .. math:: X_t[f] = \sum_{n = 0}^{N-1} \text{window}[n]\ x[t \times H + n]\ e^{-{2 \pi j f n}/{N}} Where: - :math:`t`: The :math:`t`-th input window. - :math:`f`: Frequency :math:`0 \leq f < \text{n_fft}` for `onesided=False`, or :math:`0 \leq f < \lfloor \text{n_fft} / 2 \rfloor + 1` for `onesided=True`. - :math:`N`: Value of `n_fft`. - :math:`H`: Value of `hop_length`. Args: x (Tensor): The input data which is a 1-dimensional or 2-dimensional Tensor with shape `[..., seq_length]`. It can be a real-valued or a complex Tensor. n_fft (int): The number of input samples to perform Fourier transform. hop_length (int|None, optional): Number of steps to advance between adjacent windows and `0 < hop_length`. Default: `None` (treated as equal to `n_fft//4`) win_length (int|None, optional): The size of window. Default: `None` (treated as equal to `n_fft`) window (Tensor|None, optional): A 1-dimensional tensor of size `win_length`. It will be center padded to length `n_fft` if `win_length < n_fft`. Default: `None` ( treated as a rectangle window with value equal to 1 of size `win_length`). center (bool, optional): Whether to pad `x` to make that the :math:`t \times hop\_length` at the center of :math:`t`-th frame. Default: `True`. pad_mode (str, optional): Choose padding pattern when `center` is `True`. See `paddle.nn.functional.pad` for all padding options. Default: `"reflect"` normalized (bool, optional): Control whether to scale the output by `1/sqrt(n_fft)`. Default: `False` onesided (bool, optional): Control whether to return half of the Fourier transform output that satisfies the conjugate symmetry condition when input is a real-valued tensor. It can not be `True` if input is a complex tensor. Default: `None` name (str|None, optional): The default value is None. Normally there is no need for user to set this property. For more information, please refer to :ref:`api_guide_Name`. Returns: The complex STFT output tensor with shape `[..., n_fft//2 + 1, num_frames]` (real-valued input and `onesided` is `True`) or `[..., n_fft, num_frames]` (`onesided` is `False`) Examples: .. code-block:: pycon >>> import paddle >>> from paddle.signal import stft >>> # real-valued input >>> x = paddle.randn([8, 48000], dtype=paddle.float64) >>> y1 = stft(x, n_fft=512) >>> print(y1.shape) paddle.Size([8, 257, 376]) >>> y2 = stft(x, n_fft=512, onesided=False) >>> print(y2.shape) paddle.Size([8, 512, 376]) >>> # complex input >>> x = paddle.randn([8, 48000], dtype=paddle.float64) + \ ... paddle.randn([8, 48000], dtype=paddle.float64)*1j >>> print(x.shape) paddle.Size([8, 48000]) >>> print(x.dtype) paddle.complex128 >>> y1 = stft(x, n_fft=512, center=False, onesided=False) >>> print(y1.shape) paddle.Size([8, 512, 372]) )r z9x should be a 1D or 2D real tensor, but got rank of x is r rNz"hop_length should be > 0, but got .rz"n_fft should be in (0, seq_length( )], but got "win_length should be in (0, n_fft(8expected a 1D window tensor of size equal to win_length(), but got window with shape r3r%rBconstantpadmode)rJreflectz5pad_mode should be "reflect" or "constant", but got "z".NLC)rLrM data_format)rr'r(r)rrBr permorthobackwardzBonesided should be False when input or window is a complex Tensor.T)rnr)normforwardonesidedr8rrVr)rWrXr8)lenr3 unsqueezer2rpaddleonesr%nn functionalrLsqueezer transposemultiplyrrr squeeze_)rn_fftr( win_lengthwindowcenterpad_mode normalizedrYr8x_rankpad_left pad_right pad_lengthx_framesrWr;s r<rrsn\F  L C6(KL  { KKN!_ >M? |1MM> 5'AGGBK' 0 \%PQ R ' z "U " ,UG< |1M "6<< A%#f+*C FzlRopvp|p|o}}~  C*CJ=@&1,&1 %%)) 9-J*    PC8*B O P aZ II $ $ KKO( % '"+ q"MH!! "Hx0H 7jD!(++( P | a== $RdDt  --Y- 'C { Q Jr=c v[USSS/S5 [UR5n U S;d SU 35eU S:XaURS5nUc[ US -5nUcUnSUs=:aU::dO S US US 35eSUs=:aU::dO SUS US 35eURSn URSn [ 5(aQUR S:wdS5eU(a XS-S-:XdSUS-S-SU S 35eOX:XdSUSU S 35eUbA[UR5S:Xa[U5U:XdSUSURS 35eOdUR[R[R4;a[RO[Rn[R"U4US9nX1:a8X- S-nX- U- n[RRRXOU/SS9nUR!/SQS9nU(aSOSnU (aU(aS 5e[#US SUS!S S"9nO>[%U5(aS#5eUS!LaUS S 2S S 2S US-S-24n['US SUS!S S"9n[R("UU5R!/SQS9n[+UUSS$9n[+[R,"[R("XD5RS5U S/S%9R!SS/S9USS$9nUc&U(aUS S 2US-US-*24nUUS-US-*nO%U(aUS-nOSnUS S 2UUU-24nUUUU-n[ 5(a;UR/5R15R35S&:a [5S'5eUU- nU S:Xa[R6"USS(9nU$))a( Inverse short-time Fourier transform (ISTFT). Reconstruct time-domain signal from the giving complex input and window tensor when nonzero overlap-add (NOLA) condition is met: .. math:: \sum_{t = -\infty}^{\infty} \text{window}^2[n - t \times H]\ \neq \ 0, \ \text{for } all \ n Where: - :math:`t`: The :math:`t`-th input window. - :math:`N`: Value of `n_fft`. - :math:`H`: Value of `hop_length`. Result of `istft` expected to be the inverse of `paddle.signal.stft`, but it is not guaranteed to reconstruct a exactly realizable time-domain signal from a STFT complex tensor which has been modified (via masking or otherwise). Therefore, `istft` gives the `[Griffin-Lim optimal estimate] `_ (optimal in a least-squares sense) for the corresponding signal. Args: x (Tensor): The input data which is a 2-dimensional or 3-dimensional **complex** Tensor with shape `[..., n_fft, num_frames]`. n_fft (int): The size of Fourier transform. hop_length (int|None, optional): Number of steps to advance between adjacent windows from time-domain signal and `0 < hop_length < win_length`. Default: `None` ( treated as equal to `n_fft//4`) win_length (int|None, optional): The size of window. Default: `None` (treated as equal to `n_fft`) window (Tensor|None, optional): A 1-dimensional tensor of size `win_length`. It will be center padded to length `n_fft` if `win_length < n_fft`. It should be a real-valued tensor if `return_complex` is False. Default: `None`(treated as a rectangle window with value equal to 1 of size `win_length`). center (bool, optional): It means that whether the time-domain signal has been center padded. Default: `True`. normalized (bool, optional): Control whether to scale the output by :math:`1/sqrt(n_{fft})`. Default: `False` onesided (bool, optional): It means that whether the input STFT tensor is a half of the conjugate symmetry STFT tensor transformed from a real-valued signal and `istft` will return a real-valued tensor when it is set to `True`. Default: `True`. length (int|None, optional): Specify the length of time-domain signal. Default: `None`( treated as the whole length of signal). return_complex (bool, optional): It means that whether the time-domain signal is real-valued. If `return_complex` is set to `True`, `onesided` should be set to `False` cause the output is complex. name (str|None, optional): The default value is None. Normally there is no need for user to set this property. For more information, please refer to :ref:`api_guide_Name`. Returns: A tensor of least squares estimation of the reconstructed signal(s) with shape `[..., seq_length]` Examples: .. code-block:: pycon >>> import numpy as np >>> import paddle >>> from paddle.signal import stft, istft >>> paddle.seed(0) >>> # STFT >>> x = paddle.randn([8, 48000], dtype=paddle.float64) >>> y = stft(x, n_fft=512) >>> print(y.shape) paddle.Size([8, 257, 376]) >>> # ISTFT >>> x_ = istft(y, n_fft=512) >>> print(x_.shape) paddle.Size([8, 48000]) >>> np.allclose(x, x_) True r complex64 complex128r)rBz>>{ zA~- =eqj1n=MMnownxxyz -$ 4UG;]^f]gghi $6<< A%#f+*C FzlRopvp|p|o}}~  C*C ww6>>6+;+;<< NN  J= E&1,&1 %%)) 9-J*    A!7jD a |Tu4Pf%% U % u !Q(%1*q.(()ATu4P //#v & 0 0  1 C  *2 C! ++oof-77:"A  )!Q)   N~ a%1*%1* 556C+UaZUaZ=IN QJEE!UUV^++,'?^//1557<<>F s    C {nnSq) Jr=)rN) rrr'r2r(r2r) _SignalAxesr8 str | Nonereturnr) rrr(r2r)rr8rrr)NNNTrNFNN)rrrer2r( int | Nonerfrrg Tensor | NonerhboolrizLiteral['reflect', 'constant']rjrrYz bool | Noner8rrr) NNNTFTNFN)rrrer2r(rrfrrgrrhrrjrrYrr{rr|rr8rrr) __future__rtypingrrr]rpaddle.frameworkrrr base.data_feederr base.layer_helperr fftr rrtensor.attributerrr__all__rr?rrr=r<rs #)  7***(%.K    z zzz  z  z  z|LPf ff&1f>Hf fX"! /8 r r rr r  r  r-rrr r rp"!  Q Q QQ Q  Q  QQQ QQ Q Qr=