A Hierarchical Subspace Model for Language-Attuned Acoustic Unit Discovery

In this work, we propose a hierarchical subspace model for acousticunit discovery. In this approach, we frame the task as one oflearning embeddings on a low-dimensional phonetic subspace, andsimultaneously specify the subspace itself as an embedding on a hyper-subspace. We train the hyper-subspace on a set of transcribedlanguages and transfer it to the target language. In the target language,we infer both the language and unit embeddings in an unsupervisedmanner, and in so doing, we simultaneously learn a subspaceof units specific to that language and the units that dwell on it.We conduct experiments on TIMIT and two low-resource languages:Mboshi and Yoruba. Results show that our model outperforms majoracoustic unit discovery techniques, both in terms of clusteringquality and segmentation accuracy.