On Robustness of Cloud Speech APIs
Akhil Mathur, Anton Isopoussu, Fahim Kawsar, Robert E. Smith, Nicholas D. Lane, Nadia Bianchi‐Berthouze · 2018
The robustness and consistency of sensory inference models under changing environmental conditions and hardware is a crucial requirement for the generalizability of recent innovative work, particularly in the field of deep learning, from the lab to the real world. We measure the extent to which current speech recognition cloud models are robust to background noise, and show that hardware variability is still a problem for real-world applicability of state-of-the-art speech recognition models.