Delayed-KD: Why Matching Every Frame Hurts Streaming Speech Recognition
Analysis by the aitrendblend editorial team · Knowledge distillation and model compression · 13 min read Streaming ASR Knowledge distillation CTC Low latency A streaming model has to commit to a token before it has heard the rest of the sentence, which is exactly what creates the emission delay this paper targets. A voice assistant […]
Delayed-KD: Why Matching Every Frame Hurts Streaming Speech Recognition Read More »

