# Speech Enhancement

For time-stamped speaker-attributed transcription of two-person smart-glasses conversations, extracting or separating each target speaker before decoding lowers error rates, and the gain grows with the speech overlap ratio on the validation set.

For time-stamped speaker-attributed transcription of two-person smart-glasses conversations, extracting or separating each target speaker before decoding lowers error rates, and the gain grows with the speech overlap ratio on the validation set.

The work supervises every intermediate depth of one causal speech enhancement model and fine-tunes its output heads so deeper outputs are never worse than shallower ones, yielding a family of static models that are more Pareto-efficient than equivalently-sized counterparts trained from scratch on the same budget (up to 0.11 higher PESQ at equivalent compute, matching the best PESQ at 30% less compute); after int8 quantization on an STM32N6 microcontroller, the dynamic enhancer lies on the same latency-quality frontier as the static models, with the policy running in 26 microseconds per frame on the companion Cortex-M55 and splitting the enhancer into separate NPU graphs adding 2.2% latency overhead.

The work supervises every intermediate depth of one causal speech enhancement model and fine-tunes its output heads so deeper outputs are never worse than shallower ones, yielding a family of static models that are more Pareto-efficient than equivalently-sized counterparts trained from scratch on the same budget (up to 0.11 higher PESQ at equivalent compute, matching the best PESQ at 30% less compute); after int8 quantization on an STM32N6 microcontroller, the dynamic enhancer lies on the same latency-quality frontier as the static models, with the policy running in 26 microseconds per frame on the companion Cortex-M55 and splitting the enhancer into separate NPU graphs adding 2.2% latency overhead.