Enhanced Loudspeaker Membrane Excursion Control Method Using Low Latency Distortion Prediction and Efficient LSTM Networks

Audio and video calling, listening to videos, music or podcasts over portable devices such as mobile, tablets, headphones and earbuds has become an integral part of our daily lives. However, small loudspeakers called micro-speakers used in such devices suffer from poor audio quality and insufficient volume output leading to poor listening experiences. Existing loudspeaker control techniques use current/voltage signals across loudspeaker terminals as input to control models, which reduce the loudspeaker membrane excursion based on preset thresholds. Due to inaccuracy in current/voltage sensing and heuristic-based tuning for parameters such as excursion thresholds, these techniques suffer from poor accuracy. These techniques focus on loudspeaker protection instead of superior audio quality leading to reduction in output loudness and undesirable change in timbre of multimedia audio signals. In this work, we propose a novel low-complex loudspeaker control method that uses a Fully-Connected Neural Network (FCNN) based distortion classifier and a long short-term memory (LSTM) neural network based loudspeaker model. Our proposed method improves sound quality and enhances computational efficiency by selectively processing frames that can cause audible distortions. The proposed FCNN-based distortion classifier achieves about 99 percent accuracy superior to other state-of-the-art classification methods, while proposed LSTM based loudspeaker modelling method outperforms existing state-of-the art methods while ensuring lower latencies on two different processors compared to state-of-the-art Neural Networks approaches. Our proposed algorithm can be easily fine-tuned and applied across loudspeakers in different devices to improve loudspeaker control and overall audio quality, since it does not utilize current/voltage sensing and heuristic hand-tuning of loudspeaker model parameters.