cdpam 0.0.1

A pip package for an improved perceptual audio metric

Contrastive learning-based Deep Perceptual Audio Metric (CDPAM) [Webpage]

Contrastive Learning For Perceptual Audio Similarity

Pranay Manocha, Zeyu Jin, Richard Zhang, Adam Finkelstein
Arxiv 2020()

This is a Pytorch implementation of our new and improved audio perceptual metric. It contains (0) minimal code to run our perceptual metric (CDPAM).

(0) Usage as a loss function

Minimal basic usage as a distance metric

Running the command below takes two audio files as input and gives the perceptual distance between the files. It should return (approx)distance = 0.1696. Some GPU's are non-deterministic, and so the distance could vary in the lsb.

Installing the metric (CDPAM - perceptual audio similarity metric)

pip install cdpam

Using the metric is as simple as:

import cdpam
loss_fn = cdpam.DPAM()
wav_ref = cdpam.load_audio('sample_audio/ref.wav')
wav_out = cdpam.load_audio('sample_audio/2.wav')

dist = loss_fn.forward(wav_ref,wav_out)