No abstract

We describe the design of a voice trigger detection system for smart speakers. In this study, we address two major challenges. The first is that the detectors are deployed in complex acoustic environments with external noise and loud playback by the device itself. Secondly, collecting training examples for a specific keyword or trigger phrase is challenging resulting in a scarcity of trigger phrase specific training data. We describe a two-stage cascaded architecture where a low-power detector is always running and listening for the trigger phrase. If a detection is made at this stage, the candidate audio segment is re-scored by larger, more complex models to verify that the segment contains the trigger phrase. In this study, we focus our attention on the architecture and design of these second-pass detectors. We start by training a general acoustic model that produces phonetic transcriptions given a large labelled training dataset. Next, we collect a much smaller dataset of examples that are challenging for the baseline system. We then use multi-task learning to train a model to simultaneously produce accurate phonetic transcriptions on the larger dataset and discriminate between true and easily confusable examples using the smaller dataset. Our results demonstrate that the proposed model reduces errors by half compared to the baseline in a range of challenging test conditions without requiring extra parameters.

No abstract

We describe the architecture of an always-on keyword spotting (KWS) system for battery-powered mobile devices used to initiate an interaction with the device. An always-available voice assistant needs a carefully designed voice keyword detector to satisfy the power and computational constraints of battery powered devices. We employ a multi-stage system that uses a low-power primary stage to decide when to run a more accurate (but more power-hungry) secondary detector. We describe a straightforward primary detector and explore variations that result in very useful reductions in computation (or increased accuracy for the same computation). By reducing the set of target labels from three to one per phone, and reducing the rate at which the acoustic model is operated, the compute rate can be reduced by a factor of six while maintaining the same accuracy.

Automatic speech transcription and speaker recognition are usually treated as separate tasks even though they are interdependent. In this study, we investigate training a single network to perform both tasks jointly. We train the network in a supervised multi-task learning setup, where the speech transcription branch of the network is trained to minimise a phonetic connectionist temporal classification (CTC) loss while the speaker recognition branch of the network is trained to label the input sequence with the correct label for the speaker. We present a large-scale empirical study where the model is trained using several thousand hours of labelled training data for each task. We evaluate the speech transcription branch of the network on a voice trigger detection task while the speaker recognition branch is evaluated on a speaker verification task. Results demonstrate that the network is able to encode both phonetic and speaker information in its learnt representations while yielding accuracies at least as good as the baseline models for each task, with the same number of parameters as the independent models.