by Alibaba · Model · Voice & audio
1.7B model run with custom kernels and rolling partial stitching to add streaming; 40-80 ms speech-to-text