o MeZ@sddlmZddlZddlmZmZddlmZmZm Z ddl m Z ddl m Z ddlmZdd lmZmZdd lmZmZd d gZdddZdddZ        ddd Z         ddd ZdS))OptionalN) is_complexis_floating_point)fft_r2cfft_c2rfft_c2c)check_variable_and_dtype)_non_static_mode) LayerHelper)_C_ops _legacy_C_ops)in_dygraph_mode_in_legacy_dygraphstftistftc Cs<|dvr td|dt|tr|dkrtd|dt|tr&|dkr.td|dtrF||j|krFtd|d |j|d d }trSt||||Strmd |d |d|f}t t |}||g|R}|St |dgd|t |fit } | jdd} | j| d}| j|d|i|||dd|id|S)am 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. Returns: The output frames tensor with shape `[..., frame_length, num_frames]` if `axis==-1`, otherwise `[num_frames, frame_length, ...]` where `num_framse = 1 + (x.shape[axis] - frame_length) // hop_length` Examples: .. code-block:: python import paddle from paddle.signal import frame # 1D x = paddle.arange(8) y0 = frame(x, frame_length=4, hop_length=2, axis=-1) # [4, 3] # [[0, 2, 4], # [1, 3, 5], # [2, 4, 6], # [3, 5, 7]] y1 = frame(x, frame_length=4, hop_length=2, axis=0) # [3, 4] # [[0, 1, 2, 3], # [2, 3, 4, 5], # [4, 5, 6, 7]] # 2D x0 = paddle.arange(16).reshape([2, 8]) y0 = frame(x0, frame_length=4, hop_length=2, axis=-1) # [2, 4, 3] # [[[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 = frame(x1, frame_length=4, hop_length=2, axis=0) # [3, 4, 2] # [[[0 , 1 ], # [2 , 3 ], # [4 , 5 ], # [6 , 7 ]], # # [4 , 5 ], # [6 , 7 ], # [8 , 9 ], # [10, 11]], # # [8 , 9 ], # [10, 11], # [12, 13], # [14, 15]]] # > 2D x0 = paddle.arange(32).reshape([2, 2, 8]) y0 = frame(x0, frame_length=4, hop_length=2, axis=-1) # [2, 2, 4, 3] x1 = paddle.arange(32).reshape([8, 2, 2]) y1 = frame(x1, frame_length=4, hop_length=2, axis=0) # [3, 4, 2, 2] rrUnexpected 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).frame frame_length hop_lengthaxisxZint32Zint64Zfloat16float32float64Zinput_param_namedtypeX)rrrOuttypeZinputsattrsZoutputs) ValueError isinstanceintr shaperr rrgetattrr r r locals input_dtype"create_variable_for_type_inference append_op) rrrrnameop_typer'opouthelperr"r6=D:\Projects\ConvertPro\env\Lib\site-packages\paddle/signal.pyr!sZM       rc Cs|dvr td|dt|tr|dkrtd|dd}tr+t|||}|StrDd|d |f}tt |}||g|R}|St |d gd |t |fit }|j d d } |j| d }|j|d|i||dd|id|S)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. 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:: python import paddle from paddle.signal import overlap_add # 2D x0 = paddle.arange(16).reshape([8, 2]) # [[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) # [10] # [0 , 2 , 5 , 9 , 13, 17, 21, 25, 13, 15] x1 = paddle.arange(16).reshape([2, 8]) # [[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) # [10] # [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) # [2, 1, 10] x1 = paddle.arange(32).reshape([2, 8, 1, 2]) y1 = overlap_add(x1, hop_length=2, axis=0) # [10, 1, 2] rrrrrr overlap_addrrrrr r!r#)rrr$r%)r(r)r*rr r8paddleZin_dynamic_moder,r r r r-r.r/r0) rrrr1r2r4r'r3r5r"r6r6r7r8s<5      r8TreflectFc  Csbt|dgddt|j} | dvsJd| | dkr"|d}|dur,t|d }|dks8Jd |d |dur>|}tr]d|krN|jd ks]nJd |jd d|d d|krg|kssnJd|d|d |durt|jdkrt||ksJd|d|jd n tj|f|jd}||kr||d} ||| } tj j j || | gdd}|r|dvsJd ||d} tj j j |d | | g|dd d }t|||d d}|jgdd}t||}|rdnd}t|r|rJdt|st|dd |d || d!}n t|dd |d | d"}|jgdd}| dkr/|d|S)#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[\omega] = \sum_{n = 0}^{N-1}% \text{window}[n]\ x[t \times H + n]\ % e^{-{2 \pi j \omega n}/{N}} Where: - :math:`t`: The :math:`t`-th input window. - :math:`\omega`: Frequency :math:`0 \leq \omega < \text{n\_fft}` for `onesided=False`, or :math:`0 \leq \omega < \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, 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, optional): The size of window. Default: `None`(treated as equal to `n_fft`) window (Tensor, 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: `True` name (str, 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:: python import paddle from paddle.signal import stft # real-valued input x = paddle.randn([8, 48000], dtype=paddle.float64) y1 = stft(x, n_fft=512) # [8, 257, 376] y2 = stft(x, n_fft=512, onesided=False) # [8, 512, 376] # complex input x = paddle.randn([8, 48000], dtype=paddle.float64) + \ paddle.randn([8, 48000], dtype=paddle.float64)*1j # [8, 48000] complex128 y1 = stft(x, n_fft=512, center=False, onesided=False) # [8, 512, 372] r)rr complex64 complex128r)rz9x should be a 1D or 2D real tensor, but got rank of x is rrNz"hop_length should be > 0, but got .rz"n_fft should be in (0, seq_length(z )], but got z"win_length should be in (0, n_fft(z8expected a 1D window tensor of size equal to win_length(z), but got window with shape r+r"r=constantpadmode)rAr:z9pad_mode should be "reflect" or "constant", but got "{}".ZNLC)rCrDZ data_format)rrrrrr=rpermorthobackwardzBonesided should be False when input or window is a complex Tensor.T)rnrnormforwardonesidedr1rrJrrKrLr1)r lenr+ unsqueezer*r r9onesr"nn functionalrCformatZsqueezer transposemultiplyrrrsqueeze_)rn_fftr win_lengthwindowcenterZpad_mode normalizedrMr1x_rankpad_left pad_rightZ pad_lengthZx_framesrKr4r6r6r7rsH                c Csbt|dddgdt|j} | dvsJd| | dkr"|d}|d ur,t|d }|d ur2|}d|kr<|ksEnJd ||d|krO|ksXnJd |||jd } |jd} tr|r|| |ddks{Jd|dd| n | |ksJd|| |d urt|jdkrt||ksJd||jn|jtj tj fvrtj ntj }tj |f|d}||kr||d}|||}tj jj|||gdd}|jgdd}|rdnd}| r|rJdt|d d |dd d}n(t|rJd|dur|d d d d d |ddf}t|d d |dd d}t||jgdd}t||d d}ttjt||d| dgdjddgd|d d}|d urp|ro|d d |d|d f}||d|d }n|rx|d}nd}|d d |||f}||||}tr|d krtd!||}| dkr|d|S)"ax 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 realizible time-domain signal from a STFT complex tensor which has been modified (via masking or otherwise). Therefore, `istft` gives the [Griffin-Lim optimal estimate](https://ieeexplore.ieee.org/document/1164317) (optimal in a least-squares sense) for the corresponding signal. Args: x (Tensor): The input data which is a 2-dimensional or 3-dimensional **complesx** Tensor with shape `[..., n_fft, num_frames]`. n_fft (int): The size of Fourier transform. hop_length (int, 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, optional): The size of window. Default: `None`(treated as equal to `n_fft`) window (Tensor, 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 `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, 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, 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:: python 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) # [8, 257, 376] # ISTFT x_ = istft(y, n_fft=512) # [8, 48000] np.allclose(x, x_) # True rr;r<r)r=z>x should be a 2D or 3D complex tensor, but got rank of x is {}r=rNr>z8hop_length should be in (0, win_length({})], but got {}.z3win_length should be in (0, n_fft({})], but got {}.rrzQfft_size should be equal to n_fft // 2 + 1({}) when onesided is True, but got {}.zIfft_size should be equal to n_fft({}) when onesided is False, but got {}.zZexpected a 1D window tensor of size equal to win_length({}), but got window with shape {}.r@rArBrErFrHrIzSonesided should be False when input(output of istft) or window is a complex Tensor.FrNzGData type of window should not be complex when return_complex is False.)rrr)rZ repeat_timesgdy=zAbort istft because Nonzero Overlap Add (NOLA) condition failed. For more information about NOLA constraint please see `scipy.signal.check_NOLA`(https://docs.scipy.org/doc/scipy/reference/generated/scipy.signal.check_NOLA.html).)r rOr+rTrPr*r r"r9rr;rrQrRrSrCrUrrrrVr8Ztileabsminitemr(rW)rrXrrYrZr[r\rMlengthZreturn_complexr1r]Zn_framesZfft_sizeZ window_dtyper^r_rKr4Zwindow_envelopstartr6r6r7rsS                  "     )rN)NNNTr:FTN) NNNTFTNFN)typingrr9Ztensor.attributerrZfftrrrZfluid.data_feederr Zfluid.frameworkr Zfluid.layer_helperr r r Zpaddle.fluid.frameworkrr__all__rr8rrr6r6r6r7sD      |X !