Add hannWindow and hammingWindow op

[Lewuathe]

Add Hann window op and hamming window op since this window function is necessary to implement stft and isftf as well as hamming window.

See: tensorflow/tfjs#1362

---

This change is 

[Lewuathe]

@nsthorat @annxingyuan That's necessary for implementing stft, isftf (tensorflow/tfjs#1362). Could you take a look when you get a chance?

[nsthorat]

Reviewed 1 of 7 files at r1, 6 of 10 files at r2, 1 of 1 files at r3.
Reviewable status: 0 of 1 approvals obtained (waiting on @Lewuathe)

---

src/backends/cpu/backend_cpu.ts, line 3409 at r3 (raw file):

    const newValues = new Float32Array(windowLength);
    for (let i = 0; i < windowLength; ++i) {
      const cosArg = (2.0 * Math.PI * i) / (windowLength+even-1);

can you make sure you have clang-format installed and working properly here? there should be spaces in this statement :)

---

src/backends/cpu/backend_cpu.ts, line 3410 at r3 (raw file):

    for (let i = 0; i < windowLength; ++i) {
      const cosArg = (2.0 * Math.PI * i) / (windowLength+even-1);
      newValues[i] = 0.54 - 0.46 * Math.cos(cosArg);

where do these constants come from? can we make them actual constants?

---

src/ops/signal_ops.ts, line 3 at r3 (raw file):

/**
 * @license
 * Copyright 2019 Google Inc. All Rights Reserved.

LLC

---

src/ops/signal_ops.ts, line 36 at r3 (raw file):

 */
function hannWindow_(windowLength: number): Tensor {
  return ENGINE.runKernel(backend => backend.hannWindow(windowLength), {});

In TensorFlow it looks like these don't exist as kernels, they just happen in python: https://github.com/tensorflow/tensorflow/blob/r1.13/tensorflow/python/ops/signal/window_ops.py

I think we should just do that here, generate all the values on the CPU.In practice I bet the windows aren't huge so I would bet that doing this on the CPU would actually be faster than spinning up a whole GPU program. You can remove all the backend / kernel logic and just put that as vanilla JS right in this file :)

[Lewuathe]

Reviewable status: 0 of 1 approvals obtained (waiting on @Lewuathe and @nsthorat)

---

src/backends/cpu/backend_cpu.ts, line 3410 at r3 (raw file):

Previously, nsthorat (Nikhil Thorat) wrote…

where do these constants come from? can we make them actual constants?

I checked Wikipedia as a reference as TensorFlow core does.
https://en.wikipedia.org/wiki/Window_function#Hann_and_Hamming_windows

0.54 is actually used in TensorFlow core too.
https://github.com/tensorflow/tensorflow/blob/65f12d778825306c31fcff5aa0d770bac1d4751e/tensorflow/python/ops/signal/window_ops.py#L80-L81

---

src/ops/signal_ops.ts, line 3 at r3 (raw file):

Previously, nsthorat (Nikhil Thorat) wrote…

LLC

Sorry, I don't understand what it means. I used the exact same license header as other files. What change do I need?

---

src/ops/signal_ops.ts, line 36 at r3 (raw file):

Previously, nsthorat (Nikhil Thorat) wrote…

In TensorFlow it looks like these don't exist as kernels, they just happen in python: https://github.com/tensorflow/tensorflow/blob/r1.13/tensorflow/python/ops/signal/window_ops.py

I think we should just do that here, generate all the values on the CPU.In practice I bet the windows aren't huge so I would bet that doing this on the CPU would actually be faster than spinning up a whole GPU program. You can remove all the backend / kernel logic and just put that as vanilla JS right in this file :)

Okay, will do.

[nsthorat]

Will merge after these, thanks for the PR!

Reviewable status: 0 of 1 approvals obtained (waiting on @Lewuathe)

---

src/ops/signal_ops.ts, line 3 at r3 (raw file):

Previously, Lewuathe (Kai Sasaki) wrote…

Sorry, I don't understand what it means. I used the exact same license header as other files. What change do I need?

oh sorry it's 2019 Google LLC now, the 2019 files should have that.

---

src/ops/signal_ops.ts, line 41 at r4 (raw file):

 * Generate a hamming window.
 *
 * See: https://en.wikipedia.org/wiki/Window_function#Hann_window

this is the link to hann :)

---

src/ops/signal_ops.ts, line 51 at r4 (raw file):

 * @doc {heading: 'Operations', subheading: 'Signal', namespace: 'signal'}
 */
function hammingWindow_(windowLength: number): Tensor {

return can be stricter, Tensor1D here and above

[nsthorat]

Reviewed 1 of 1 files at r5.
Reviewable status: complete! 1 of 1 approvals obtained

[nsthorat]

Reviewable status: complete! 1 of 1 approvals obtained (waiting on @Lewuathe)

---

src/ops/signal_ops.ts, line 62 at r5 (raw file):

    newValues[i] = a - b * Math.cos(cosArg);
  }
  return Tensor.make([windowLength], {values: newValues}) as Tensor1D;

okay I promise this is the last comment, you should use the public API here, just call the tensor1d(values, [windowLength]) function

[Lewuathe]

Thanks. Updated.

[nsthorat]

Reviewed 1 of 1 files at r6.
Reviewable status: complete! 1 of 1 approvals obtained (waiting on @Lewuathe)

[nsthorat]

Thanks for the PR!