Towards efficient models for real-time deep noise suppression
This work addresses the problem of resource efficiency in real-time deep noise suppression for applications like speech enhancement, but it is incremental as it focuses on optimizing existing architectures rather than introducing a new paradigm.
The paper tackled the challenge of developing compact deep learning models for real-time speech enhancement by investigating small recurrent and convolutional-recurrent architectures trained on a large dataset with reverberation, showing tradeoffs between computational complexity and speech quality measured on real recordings using a MOS estimator.
With recent research advancements, deep learning models are becoming attractive and powerful choices for speech enhancement in real-time applications. While state-of-the-art models can achieve outstanding results in terms of speech quality and background noise reduction, the main challenge is to obtain compact enough models, which are resource efficient during inference time. An important but often neglected aspect for data-driven methods is that results can be only convincing when tested on real-world data and evaluated with useful metrics. In this work, we investigate reasonably small recurrent and convolutional-recurrent network architectures for speech enhancement, trained on a large dataset considering also reverberation. We show interesting tradeoffs between computational complexity and the achievable speech quality, measured on real recordings using a highly accurate MOS estimator. It is shown that the achievable speech quality is a function of network complexity, and show which models have better tradeoffs.