Skip to main content

A Conv-STFT/iSTFT implement based on Torch

Project description

Conv-STFT/iSTFT in PyTorch

Author: Shimin Zhang

The code refers to the following repo:

  1. remove modulation effects
  2. enframe and conv-overlap-add

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


Install easily with pip:pip install conv_stft or download this repo, python install.


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 =

stft = STFT(

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



Test it by just cloning this repo and running

pip install -r requirements.txt
python -m pytest .

Project details

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Files for conv-stft, version 0.2.0
Filename, size File type Python version Upload date Hashes
Filename, size conv_stft-0.2.0-py3.6.egg (8.4 kB) File type Egg Python version 0.2.0 Upload date Hashes View

Supported by

AWS AWS Cloud computing Datadog Datadog Monitoring DigiCert DigiCert EV certificate Facebook / Instagram Facebook / Instagram PSF Sponsor Fastly Fastly CDN Google Google Object Storage and Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Salesforce Salesforce PSF Sponsor Sentry Sentry Error logging StatusPage StatusPage Status page