morenolaquatra / auau

(Au)dio (Au)gmentation = auau

Hi, I'm Moreno La Quatra

I'm a fellow researcher at STMLab.

• I’m interested in 📝 NLP, 🔉 Audio processing and 🖼️ Multimodal Learning.
• If you are a master student 👨‍🎓 👩‍🎓 interested in some of the fields above, you may want to take a look to this repo.
• I'm looking to collaborate on the analysis of spoken content 🗣️, get in touch if you are interested!
• 📫 Reach me by mail or on Twitter.

auau's Issues

[Code refactoring] Move all Augmenter methods to their own classes

I have yet to take a closer look at all of the augmenting functions, but it seems to be that they would benefit from a refactoring where they are assigned to a separate class that extends a base class Augmenter. Something along the lines of:

class WhiteNoise(Augmenter):
    def __init__(self, noise_factor=0.25):
        ...
    def __call__(self, signal):
        """ Adding white noise, processing on self.device
        """
        signal = self._tensor_on_device(signal)
        signal = self._check_and_convert_mono(signal)
        noise_signal = torch.normal(0, signal.std(), signal.size()).to(self.device)
        augmented_signal = signal + noise_signal * self.noise_factor
        return augmented_signal

To be then used as:

wn = WhiteNoise(noise_factor=0.1)
wn(signal)

This, on top of being (imho) cleaner, also paves the way for a pipeline-oriented implementation:

pipeline = SomePipelineClass([WhiteNoise(...), Resample(...), LowpassFilter(...)])
pipeline(signal)