Moshi
Emerging1papers using it
2026first seen
The 'Moshi' dataset is a benchmark that contains real-time, full-duplex speech interaction samples used to evaluate the performance of speech language models in handling natural conversational phenomena such as overlap and interruption.