MSRN and Multi-Headed Attention Mechanism for Language Identification

With the popularity of mobile internet, people all over world can easily create and publish diverse media content such as multilingual multi-dialectal audio video. Therefore, language or dialect identification (LID) is increasingly important for practical applications cross lingual processing front-end part subsequent tasks speech recognition voice identification. This paper proposes a neural network framework based on multiscale residual (MSRN) multi-headed self-attention (MHSA). Experimental results show that this method effectively improve accuracy robustness compared to other methods. model uses MSRN extract spectrogram feature MHSA filter useful features suppress irrelevant features. Training test sets are constructed from both “Common Voice” “Oriental Language Recognition” (AP17-OLR) datasets. The experimental LID.