Stft

English | 简体中文

Product Support Status

Product	Supported
Ascend 950PR&950DT products	√
Atlas A3 products	√
Atlas A2 products	√
Atlas 200I/500 A2 inference products	×
Atlas inference products	×
Atlas training products	×

Function Description

• Operator function: computes the Fourier transform of the input within a sliding window.

• Computation formula:

  • When normalized=false:

    $$X[w,m]=\sum _{k=0}^{winLength-1}window[k]\ast self[m\ast hopLength+k]\ast exp(-j\ast \frac{2\pi wk}{nFft})$$

  • When normalized=true:

    $$X[w,m]=\frac{1}{\sqrt{nFft}}(\sum _{k=0}^{winLength-1}window[k]\ast self[m\ast hopLength+k]\ast exp(-j\ast \frac{2\pi wk}{nFft}))$$

  Here:

  • $w$ is the frequency point of the FFT.
  • $m$ is the index of the sliding window.
  • $self$ is a 1D or 2D Tensor. If $self$ is 1D, it is a time-series sampling sequence. If $self$ is 2D, it is multiple time-series sampling sequences.
  • $hopLength$ is the sliding window size.
  • $window$ is a 1D Tensor and the window function of the STFT (for example, hann_window), whose length is $winLength$.
  • $exp(-j\ast \frac{2\pi wk}{nFft})$ is the twiddle factor.

Parameter Description

Parameter Name	Input/Output/Attribute	Description	Data Type	Data Format
x	Input	Input to be computed, corresponding to `self` in the formula. It must be a 1D/2D Tensor with the shape [L]/[B, L], where L is the length of the time-series sampling sequence and B is the number of time-series sampling sequences. Inputs Inf, -Inf, and NaN are not supported.	FLOAT32, DOUBLE, COMPLEX64, COMPLEX128	ND
window	Optional input	It must be a 1D Tensor, corresponding to `window` in the formula. The shape is [winLength], where winLength is the length of the STFT window function, and the data type is the same as that of `x`.	FLOAT32, DOUBLE, COMPLEX64, COMPLEX128	ND
hop_length	Optional attribute	Spacing between sliding windows (greater than or equal to 0), corresponding to `hopLength` in the formula. The default value is n_fft/4.	INT64	-
win_length	Optional attribute	Size of window (greater than or equal to 0), corresponding to `winLength` in the formula. The default value is n_fft.	INT64	-
normalized	Optional attribute	Whether to normalize the Fourier transform result. The default value is false.	BOOL	-
onesided	Optional attribute	Whether to return the full result or half of the result. The default value is true.	BOOL	-
return_complex	Optional attribute	Whether the return value is a complex Tensor or a Tensor with the real and imaginary parts separated. The default value is true.	BOOL	-
n_fft	Attribute	Number of FFT points (greater than 0), corresponding to `nFft` in the formula.	INT64	-
y	Output	Result of the Fourier transform of `x` within `window`. It must be a 2D/3D/4D Tensor, corresponding to `X[w,m]` in the formula. If return_complex=True, y is a complex Tensor with the shape [N, T] or [B, N, T]. If return_complex=False, y is a real Tensor with the shape [N, T, 2] or [B, N, T, 2]. Here, N=n_fft(onesided=False) or (n_fft // 2 + 1)(onesided=True); T is the number of sliding windows, T = (L - n_fft) // hop_length + 1.	FLOAT32, DOUBLE, COMPLEX64, COMPLEX128	ND

Constraint Description

• n_fft <= L.
• win_length <= n_fft.

Invocation Description

Invocation Method	Sample Code	Description
aclnn API	test_aclnn_stft	Invokes the Stft operator through the aclStft API.
Graph mode	-	Invokes the Stft operator by building a graph through the operator IR.