istftLayer

Inverse short-time Fourier transform layer

Since R2024a

Description

An ISTFT layer computes the inverse short-time Fourier transform of the input. Use of this layer requires Deep Learning Toolbox™.

Creation

Syntax

layer = istftLayer

layer = istftLayer(Name=Value)

Description

layer = istftLayer creates an Inverse Short-Time Fourier Transform (ISTFT) layer. The input to istftLayer must be a real-valued dlarray (Deep Learning Toolbox) object in "CBT" or "SCBT" format.

For "CBT" inputs, the size of the channel ("C") dimension must be even and divisible by floor(FFTLength/2)+1.
For "SCBT" inputs, the size of the spatial ("S") dimension must equal floor(FFTLength/2)+1.

The output of istftLayer is a real-valued array in "CBT" format.

For more information, see Layer Input and Output Formats.

example

layer = istftLayer(Name=Value) creates an ISTFT layer with properties specified by one or more name-value arguments. You can specify the analysis window and the number of overlapped samples, among others.

Properties

expand all

ISTFT

`Window` — Windowing function
`hann(128,"periodic")` (default) | vector

This property is read-only.

Windowing function used to compute the ISTFT, specified as a vector with two or more elements. For perfect reconstruction, use the same window as in stftLayer.

For a list of available windows, see Windows.

You can set this property when you create an istftLayer object. After you create an istftLayer object, this property is read-only.

Note

istftLayer initializes the weights internally so that Window is in single precision. Initializing the weights directly is not recommended.

Example: hann(N+1) and (1-cos(2*pi*(0:N)'/N))/2 both specify a Hann window of length N + 1.

Data Types: double | single

`OverlapLength` — Number of overlapped samples
`75%` of window length (default) | nonnegative integer

This property is read-only.

Number of overlapped samples, specified as a nonnegative integer smaller than the length of window. If you omit OverlapLength or specify it as empty, the object sets it to the largest integer less than 75% of the window length, which turns out to be 96 samples for the default Hann window. Equivalently, the stride between adjoining segments is 32 samples.

You can set this property when you create an istftLayer object. After you create an istftLayer object, this property is read-only.

Data Types: double | single

`FFTLength` — Number of DFT points
`128` (default) | positive integer

This property is read-only.

Number of discrete Fourier transform (DFT) points, specified as a positive integer greater than or equal to the window length. To achieve perfect time-domain reconstruction, set the number of DFT points to match that used in stftLayer.

You can set this property when you create an istftLayer object. After you create an istftLayer object, this property is read-only.

Data Types: double | single

`Method` — Method of overlap-add
`"wola"` (default) | `"ola"`

This property is read-only.

Method of overlap-add, specified as one of these:

"wola" — Weighted overlap-add
"ola" — Overlap-add

You can set this property when you create an istftLayer object. After you create an istftLayer object, this property is read-only.

`ExpectedOutputSize` — Expected number of channels and samples
`"none"` (default) | two-element vector

This property is read-only.

Expected number of channels and samples output by istftLayer, specified as a two-element vector of positive integers. The first element is the expected number of channels, and the second element is the expected number of time samples.

By default, istftLayer does not check the output size of the ISTFT. If you specify ExpectedOutputSize, istftLayer errors if the inverse short-time Fourier transform for the given inputs do not match ExpectedOutputSize in the number of channels and samples.

You can set this property when you create an istftLayer object. After you create an istftLayer object, this property is read-only.

Data Types: single | double

Layer

`WeightLearnRateFactor` — Multiplier for weight learning rate
`0` (default) | nonnegative scalar

Multiplier for weight learning rate, specified as a nonnegative scalar. If you do not specify this property, it defaults to zero, resulting in weights that do not update with training. You can also set this property using the setLearnRateFactor (Deep Learning Toolbox) function.

Data Types: double | single

`Name` — Layer name
`""` (default) | character vector | string scalar

Layer name, specified as a character vector or a string scalar. For Layer array input, the trainnet (Deep Learning Toolbox) and dlnetwork (Deep Learning Toolbox) functions automatically assign names to layers with the name "".

The istftLayer object stores this property as a character vector.

Data Types: char | string

`NumInputs` — Number of inputs
`1` (default)

This property is read-only.

Number of inputs to the layer, returned as 1. This layer accepts a single input only.

Data Types: double

`InputNames` — Input names
`{'in'}` (default)

This property is read-only.

Input names, returned as {'in'}. This layer accepts a single input only.

Data Types: cell

`NumOutputs` — Number of outputs
`1` (default)

This property is read-only.

Number of outputs from the layer, returned as 1. This layer has a single output only.

Data Types: double

`OutputNames` — Output names
`{'out'}` (default)

This property is read-only.

Output names, returned as {'out'}. This layer has a single output only.

Data Types: cell

Examples

collapse all

Create ISTFT Layer

This example uses:

Open Live Script

Create an inverse short-time Fourier transform layer. Specify a 64-sample Hamming window. Specify 63 overlapped samples between adjoining segments.

layer = istftLayer(Window=hamming(64),OverlapLength=63)

layer = 
  istftLayer with properties:

                     Name: ''
    WeightLearnRateFactor: 0
                   Window: [64x1 double]
            OverlapLength: 63
                FFTLength: 64
                   Method: 'wola'
       ExpectedOutputSize: 'none'

   Learnable Parameters
                  Weights: [64x1 single]

   State Parameters
    No properties.

Use properties method to see a list of all properties.

Using `istftLayer` in Deep Learning Network

This example uses:

Open Live Script

Create an array of five layers, containing a sequence input layer, an STFT layer, an LSTM layer, an ISTFT layer, and a regression layer. There is one feature in the sequence input. Set the minimum signal length in the sequence input layer to 2048 samples. Use the default window of length 128 for both STFT and ISTFT layers.

layers = [
    sequenceInputLayer(1,MinLength=2048)
    stftLayer(TransformMode="realimag")
    lstmLayer(130)
    fullyConnectedLayer(130)
    istftLayer];

Create a random array containing a batch of 10 signals and 2048 samples. Save the signal as a dlarray in "CBT" format. Analyze the layers as a dlnetwork using the example network input.

networkInput = dlarray(randn(1,10,2048,"single"),"CBT");
analyzeNetwork(layers,networkInput,targetusage="dlnetwork")

ISTFT of Zero-Padded Data in Deep Learning Network

This example uses:

Open Live Script

Create a deep learning network that demonstrates perfect reconstruction of the short-time Fourier transform (STFT) of a dlarray. To minimize edge effects, the network zero-pads the data before computing the STFT.

Generate a 3-by-2000-by-5 array containing five batches of a three-channel sinusoidal signal sampled at 1 kHz for two seconds. Save the array as a dlarray, specifying the dimensions in order. dlarray permutes the array dimensions to the "CBT" shape expected by a deep learning network. Display the array dimension sizes.

Fs = 1e3;
nchan = 3;
nbtch = 5;
nsamp = 2000;
t = (0:nsamp-1)/Fs;

x = zeros(nchan,nsamp,nbtch);
for k=1:nbtch
    x(:,:,k) = sin(k*pi.*(1:nchan)'*t)+cos(k*pi.*(1:nchan)'*t);
end

xd = dlarray(x,"CTB");

Design a periodic Hann window of length 100 and set the number of overlap samples to 75. Check the window and overlap length for COLA compliance.

nwin = 100;
win = hann(nwin,"periodic");
noverlap = 75;

tf = iscola(win,noverlap)

tf = logical
   1

Create a STFT layer that uses the Hann window. Set the number of overlap samples to 75 and FFT length to 128. Set the layer transform mode to "realimag" to concatenate the real and imaginary parts of the layer output along the channel dimension. Create an ISTFT layer using the same FFT length, window, and overlap.

fftlen = 128;

ftl = stftLayer(Window=win,FFTLength=fftlen, ...
    OverlapLength=noverlap,TransformMode="realimag");

iftl = istftLayer(Window=win,FFTLength=fftlen, ...
    OverlapLength=noverlap);

Create a deep learning network appropriate for the data that demonstrates perfect reconstruction of the STFT. Use a function layer to zero-pad the data on both sides along the time dimension before computing the STFT. The length of the zero-pad is the window length. Use a function layer after the ISTFT layer to trim both sides of the ISTFT layer output by the same amount.

layers = [
    sequenceInputLayer(nchan,MinLength=nsamp)
    functionLayer(@(X) paddata(X,nsamp+2*nwin,dimension=3,side="both"))
    ftl
    iftl
    functionLayer(@(X) trimdata(X,nsamp,dimension=3,side="both"))];
dlnet = dlnetwork(layers);

Analyze the network using the data. The number of channels of the STFT layer output is twice the layer input.

analyzeNetwork(dlnet,xd)

Run the data through the forward method of the network.

dataout = forward(dlnet,xd);

The output is a dlarray in "CBT" format. Convert the network output to a numeric array. Permute the dimensions so that each page is a batch.

xrec = extractdata(dataout);
xrec = permute(xrec,[1 3 2]);

Choose a batch. Plot the original and reconstructed multichannel signal of that batch as a stacked plot.

wb = 4;
tiledlayout(2,1)
nexttile
stackedplot(x(:,:,wb)',DisplayLabels="Channel "+string(1:nchan))
title("Batch "+num2str(wb)+": Original")
nexttile
stackedplot(xrec(:,:,wb)',DisplayLabels="Channel "+string(1:nchan))
title("Batch "+num2str(wb)+": Reconstruction")

Confirm perfect reconstruction of the data.

max(abs(x(:)-xrec(:)))

ans = single
    6.2170e-07

More About

expand all

Inverse Short-Time Fourier Transform

The inverse short-time Fourier transform is computed by taking the IFFT of each DFT vector of the STFT and overlap-adding the inverted signals. The ISTFT is calculated as follows.

$\begin{matrix} x (n) = \int_{- 1 / 2}^{1 / 2} \sum_{m = - \infty}^{\infty} X_{m} (f) e^{j 2 π f n} d f \\ = \sum_{m = - \infty}^{\infty} \int_{- 1 / 2}^{1 / 2} X_{m} (f) e^{j 2 π f n} d f \\ = \sum_{m = - \infty}^{\infty} x_{m} (n), \end{matrix}$

where $R$ is the hop size between successive DFTs, $X_{m}$ is the DFT of the windowed data centered about time $m R,$ and $x_{m} (n) = x (n) g (n - m R)$ . The inverse STFT is a perfect reconstruction of the original signal as long as $\sum_{m = - \infty}^{\infty} g^{a + 1} (n - m R) = c \forall n \in ℤ,$ where the analysis window $g (n)$ was used to window the original signal and $c$ is a constant. This figure depicts the steps in reconstructing the original signal.

Constant Overlap-Add (COLA) Constraint

To ensure successful reconstruction of nonmodified spectra, the analysis window must satisfy the COLA constraint. In general, if the analysis window satisfies the condition $\sum_{m = - \infty}^{\infty} g^{a + 1} (n - m R) = c \forall n \in ℤ,$ , the window is considered to be COLA-compliant. Additionally, COLA compliance can be described as either weak or strong.

Weak COLA compliance implies that the Fourier transform of the analysis window has zeros at frame-rate harmonics such that

$G (f_{k}) = 0, k = 1, 2, \dots, R - 1, f_{k} ≜ \frac{k}{R} .$
Alias cancellation is disturbed by spectral modifications. Weak COLA relies on alias cancellation in the frequency domain. Therefore, perfect reconstruction is possible using weakly COLA-compliant windows as long as the signal has not undergone any spectral modifications.
For strong COLA compliance, the Fourier transform of the window must be bandlimited consistently with downsampling by the frame rate such that

$G (f) = 0, f \geq \frac{1}{2 R} .$
This equation shows that no aliasing is allowed by the strong COLA constraint. Additionally, for strong COLA compliance, the value of the constant $c$ must equal 1. In general, if the short-time spectrum is modified in any way, a stronger COLA compliant window is preferred.

You can use the iscola function to check for weak COLA compliance. The number of summations used to check COLA compliance is dictated by the window length and hop size. In general, it is common to use $a = 1$ in $\sum_{m = - \infty}^{\infty} g^{a + 1} (n - m R) = c \forall n \in ℤ,$ for weighted overlap-add (WOLA), and $a = 0$ for overlap-add (OLA). By default, istft uses the WOLA method, by applying a synthesis window before performing the overlap-add method.

In general, the synthesis window is the same as the analysis window. You can construct useful WOLA windows by taking the square root of a strong OLA window. You can use this method for all nonnegative OLA windows. For example, the root-Hann window is a good example of a WOLA window.

Perfect Reconstruction

In general, computing the STFT of an input signal and inverting it does not result in perfect reconstruction. If you want the output of ISTFT to match the original input signal as closely as possible, the signal and the window must satisfy the following conditions:

Input size — If you invert the output of stft using istft and want the result to be the same length as the input signal x, the value of

$k = \frac{N_{x} - L}{M - L}$
must be an integer. In the equation, N_x is the length of the signal, M is the length of the window, and L is the overlap length.
COLA compliance — Use COLA-compliant windows, assuming that you have not modified the short-time Fourier transform of the signal.
Padding — If the length of the input signal is such that the value of k is not an integer, zero-pad the signal before computing the short-time Fourier transform. Remove the extra zeros after inverting the signal.

Layer Input and Output Formats

The input to istftLayer must be a real-valued dlarray object in "CBT" or "SCBT" format. The output is a real-valued "CBT" array.

If the input is in "SCBT" format, the "S" dimension corresponds to frequency.

If the input is in "CBT" format, istftLayer assumes the frequency and channel dimensions have been flattened into the channel dimension.

Algorithms

The size of the ISTFT depends on the dimensions and data format of the input STFT, the length of the windowing function, the number of overlapped samples and the number of DFT points.

Define the hop size as hopSize = length(Window)-OverlapLength. The number of samples in the ISTFT is length(Window)+(nseg-1)*hopSize, where nseg is the size of the input in the time ("T") dimension.

If the input to istftLayer is a "SCBT" formatted dlarray, the number of channels is szC/2, where szC is the size of the input in the channel ("C") dimension.
If the input to istftLayer is a "CBT" formatted dlarray, the number of channels is szC/(2*nfreq), where nfreq = floor(FFTLength/2)+1.

Version History

Introduced in R2024a

istftLayer

Description

Creation

Syntax

Description

Properties

ISTFT

`Window` — Windowing function
`hann(128,"periodic")` (default) | vector

`OverlapLength` — Number of overlapped samples
`75%` of window length (default) | nonnegative integer

`FFTLength` — Number of DFT points
`128` (default) | positive integer

`Method` — Method of overlap-add
`"wola"` (default) | `"ola"`

`ExpectedOutputSize` — Expected number of channels and samples
`"none"` (default) | two-element vector

Layer

`WeightLearnRateFactor` — Multiplier for weight learning rate
`0` (default) | nonnegative scalar

`Name` — Layer name
`""` (default) | character vector | string scalar

`NumInputs` — Number of inputs
`1` (default)

`InputNames` — Input names
`{'in'}` (default)

`NumOutputs` — Number of outputs
`1` (default)

`OutputNames` — Output names
`{'out'}` (default)

Examples

Create ISTFT Layer

Using `istftLayer` in Deep Learning Network

ISTFT of Zero-Padded Data in Deep Learning Network

More About

Inverse Short-Time Fourier Transform

Constant Overlap-Add (COLA) Constraint

Perfect Reconstruction

Layer Input and Output Formats

Algorithms

Version History

See Also

Apps

Objects

Functions

Topics

istftLayer

Description

Creation

Syntax

Description

Properties

ISTFT

Window — Windowing function hann(128,"periodic") (default) | vector

OverlapLength — Number of overlapped samples 75% of window length (default) | nonnegative integer

FFTLength — Number of DFT points 128 (default) | positive integer

Method — Method of overlap-add "wola" (default) | "ola"

ExpectedOutputSize — Expected number of channels and samples "none" (default) | two-element vector

Layer

WeightLearnRateFactor — Multiplier for weight learning rate 0 (default) | nonnegative scalar

Name — Layer name "" (default) | character vector | string scalar

NumInputs — Number of inputs 1 (default)

InputNames — Input names {'in'} (default)

NumOutputs — Number of outputs 1 (default)

OutputNames — Output names {'out'} (default)

Examples

Create ISTFT Layer

Using istftLayer in Deep Learning Network

ISTFT of Zero-Padded Data in Deep Learning Network

More About

Inverse Short-Time Fourier Transform

Constant Overlap-Add (COLA) Constraint

Perfect Reconstruction

Layer Input and Output Formats

Algorithms

Version History

See Also

Apps

Objects

Functions

Topics

`Window` — Windowing function
`hann(128,"periodic")` (default) | vector

`OverlapLength` — Number of overlapped samples
`75%` of window length (default) | nonnegative integer

`FFTLength` — Number of DFT points
`128` (default) | positive integer

`Method` — Method of overlap-add
`"wola"` (default) | `"ola"`

`ExpectedOutputSize` — Expected number of channels and samples
`"none"` (default) | two-element vector

`WeightLearnRateFactor` — Multiplier for weight learning rate
`0` (default) | nonnegative scalar

`Name` — Layer name
`""` (default) | character vector | string scalar

`NumInputs` — Number of inputs
`1` (default)

`InputNames` — Input names
`{'in'}` (default)

`NumOutputs` — Number of outputs
`1` (default)

`OutputNames` — Output names
`{'out'}` (default)

Using `istftLayer` in Deep Learning Network