How are you handling the on device speech pipeline, especially around model size, latency, and accuracy tradeoffs on consumer hardware?
How are you handling the on device speech pipeline, especially around model size, latency, and accuracy tradeoffs on consumer hardware?