TurboBias 2.0: Streaming Context-Biasing for Production-Efficient ASR Systems
TurboBias 2.0 targets phrase boosting that can run per user inside batched, streaming ASR.
The paper extends GPU-accelerated TurboBias with a case-insensitive boosting graph and per-stream batched decoding. That lets each utterance in a batch use its own context list, avoiding shared or mixed personalization across simultaneous users. The authors say it works for offline and streaming inference, with greedy or beam-search decoding. Experiments report better contextual phrase recognition while keeping latency and throughput low. ArXiv · AI/CL/LG's note
The paper extends GPU-accelerated TurboBias with a case-insensitive boosting graph and per-stream batched decoding. That lets each utterance in a batch use its own context list, avoiding shared or mixed personalization across simultaneous users. The authors say it works for offline and streaming inference, with greedy or beam-search decoding. Experiments report better contextual phrase recognition while keeping latency and throughput low. ArXiv · AI/CL/LG's note
score 4