StreamHear: Domain-Adapted Pseudo-Labeling for Semi-Supervised Streaming Speech Recognition
StreamHear optimizes streaming ASR through semi-supervised pseudo-labeling and prior-regularized realignment.
Streaming automatic speech recognition (ASR) underperforms on domain-shifted target audio, where labeled in-domain data is costly to prepare while unlabeled audio is abundant. We present StreamHear, a semi-supervised pipeline that adapts a pretrained streaming student by fine-tun...
ArXiv CS.CL
Source
Generated insight by OnAirToday AI
StreamHear demonstrates that combining offline teacher-student distillation with dynamic alignment can significantly bridge the performance gap in real-time speech systems. This approach unlocks the use of massive unlabeled audio pools, making high-quality ASR accessible for specialized industries with limited training resources.
Comments
Keep comments relevant and respectful. No links, advertising, self-promotion, or spam.