Fine-tune the Pretrained ATST Model for Sound Event Detection
Fine-tuning ATST-Frame for sound event detection and achieving new SOTA results on DESED development set.
Welcome to my site!
I am a PhD. student @ Westlake University. My research interests include sound event detection ; semi-supervised / self-supervised learning in audio processing.
Feel free to contact me: sao_year@126.com
I am a PhD. student @ Westlake University. My research interests include sound event detection ; semi-supervised / self-supervised learning in audio processing.
Feel free to contact me: sao_year@126.com
Publications
Frame-wise Streaming End-to-end Speaker Diarization with Non-autoregressive Self-attention-based Attractors
A low-latency online EEND framework with causal encoding and non-autoregressive attractor decoding.
Self-supervised Audio Teacher-Student Transformer for Both Clip-level and Frame-level Tasks
ATST-Clip and ATST-Frame for unified clip-level and frame-level tasks with broad SOTA gains.
ATST Self-supervised plus RCT Semi-supervised Sound Event Detection: Submission to DCASE 2022 Challenge Task 4
Integrating ATST-Clip with RCT-CRNN, ranking 4th in DCASE 2022 Task 4 single-model setting.
RCT: Random Consistency Training for Sound Event Detection
Random Consistency Training for SED with clear improvements across CRNN-based systems.
Mixhead: Breaking the low-rank bottleneck in multi-head attention language models
Learnable head-mixing to mitigate low-rank bottlenecks in attention for language modeling and GLUE.