Figure 2 - available via license: Creative Commons Attribution 4.0 International
Content may be subject to copyright.
Source publication
We investigate robustness properties of pre-trained neural models for automatic speech recognition. Real life data in machine learning is usually very noisy and almost never clean, which can be attributed to various factors depending on the domain, e.g. outliers, random noise and adversarial noise. Therefore, the models we develop for various tasks...
Context in source publication
Context 1
... perturbation speeds up or slows down the speech. For a given speech signal x and speed 100/ρ, f (x) is computed by resampling the audio signal without changing the sampling rate, using the technique in [21]. See Fig. 2 for the results. The plot conforms with our intuition that speech that is sped up or slowed ...