https://github.com/echocatzh/conv-stft
A STFT/iSTFT written up in PyTorch using 1D Convolutions
https://github.com/echocatzh/conv-stft
conv-stft padding stft
Last synced: 13 days ago
JSON representation
A STFT/iSTFT written up in PyTorch using 1D Convolutions
- Host: GitHub
- URL: https://github.com/echocatzh/conv-stft
- Owner: echocatzh
- License: mit
- Created: 2020-08-17T18:38:08.000Z (over 5 years ago)
- Default Branch: master
- Last Pushed: 2024-07-09T17:55:12.000Z (over 1 year ago)
- Last Synced: 2026-01-06T18:12:52.534Z (20 days ago)
- Topics: conv-stft, padding, stft
- Language: Python
- Homepage:
- Size: 188 KB
- Stars: 32
- Watchers: 2
- Forks: 11
- Open Issues: 1
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# Conv-STFT/iSTFT in PyTorch
NOTE!!!: This package will no longer be maintained, and the API calls in this repo are exactly the same in [torch-mfcc](https://github.com/echocatzh/torch-mfcc).
```python
from conv_stft import STFT # same as below
from torch_mfcc import STFT
```
Author: Shimin Zhang
The code refers to the following repo:
1. [remove modulation effects](https://github.com/pseeth/torch-stft)
2. [enframe and conv-overlap-add](https://github.com/huyanxin/phasen/blob/master/model/conv_stft.py)
An STFT/iSTFT written up in PyTorch(py3) using 1D Convolutions. There are two window logic, `break` and `continue`.
- `break` - a kaldi-like framing method
When the parameters `win_len` and `fft_len` are different, padding `fft_len`-`win_len` zero points after each frame( len(frame) = `win_len` ), and the window ( len(window) = `win_len` ) always wise-multiply with frame before padding.
- `continue` - a librosa-like framing method.
When the parameters `win_len` and `fft_len` are different, framing the signal using `win_len`=`fft_len`, and zero padding on both sides of window ( len(window) = `win_len` ), which is `len(center_pad(window))=fft_len`
## Installation
Install easily with pip:`pip install conv_stft` or download this repo, `python setup.py install`.
## Usage
```python3
import torch
from conv_stft import STFT
import numpy as np
import librosa
import matplotlib.pyplot as plt
audio = librosa.load(librosa.util.example_audio_file(), duration=10.0, offset=30)[0]
device = 'cpu'
fft_len = 1024
win_hop = 256
win_len = 1024
window = 'hann'
audio = torch.FloatTensor(audio)
audio = audio.unsqueeze(0)
audio = audio.to(device)
stft = STFT(
fft_len=fft_len,
win_hop=win_hop,
win_len=win_len,
win_type=window,
).to(device)
magnitude, phase = stft.transform(audio, return_type='magphase') # 'magphase' or 'realimag'
output = stft.inverse(magnitude, phase, input_type='magphase') # 'magphase' or 'realimag'
output = output.cpu().data.numpy()[..., :]
audio = audio.cpu().data.numpy()[..., :]
print(np.mean((output - audio) ** 2)) # on order of 1e-15
```
Output of [`compare_stft.py`](compare_stft.py):
## Tests
Test it by just cloning this repo and running
```
pip install -r requirements.txt
python -m pytest .
```