Distilled Non-Semantic Speech Embeddings with Binary Neural Networks for Low-Resource Devices
This enables efficient non-semantic speech processing on wearables and other low-resource devices, though it is incremental as it builds on existing distillation and binary network techniques.
The authors tackled the problem of deploying speech models on low-resource devices by introducing BRILLsson, a binary neural network trained via knowledge distillation, which achieves comparable performance to large models while being only 2MB in size with under 8ms latency.
This work introduces BRILLsson, a novel binary neural network-based representation learning model for a broad range of non-semantic speech tasks. We train the model with knowledge distillation from a large and real-valued TRILLsson model with only a fraction of the dataset used to train TRILLsson. The resulting BRILLsson models are only 2MB in size with a latency less than 8ms, making them suitable for deployment in low-resource devices such as wearables. We evaluate BRILLsson on eight benchmark tasks (including but not limited to spoken language identification, emotion recognition, health condition diagnosis, and keyword spotting), and demonstrate that our proposed ultra-light and low-latency models perform as well as large-scale models.