MUST: A Multilingual Student-Teacher Learning approach for low-resource speech recognition
This addresses data scarcity for low-resource language speech recognition, though it is incremental as it builds on existing knowledge distillation methods.
The paper tackles the limitation of knowledge distillation for low-resource speech recognition by proposing MUST, a multilingual student-teacher learning approach that uses posteriors mapping to transfer knowledge across languages with different character sets, resulting in up to 9.5% relative reduction in character error rate compared to a monolingual baseline.
Student-teacher learning or knowledge distillation (KD) has been previously used to address data scarcity issue for training of speech recognition (ASR) systems. However, a limitation of KD training is that the student model classes must be a proper or improper subset of the teacher model classes. It prevents distillation from even acoustically similar languages if the character sets are not same. In this work, the aforementioned limitation is addressed by proposing a MUltilingual Student-Teacher (MUST) learning which exploits a posteriors mapping approach. A pre-trained mapping model is used to map posteriors from a teacher language to the student language ASR. These mapped posteriors are used as soft labels for KD learning. Various teacher ensemble schemes are experimented to train an ASR model for low-resource languages. A model trained with MUST learning reduces relative character error rate (CER) up to 9.5% in comparison with a baseline monolingual ASR.