SenseVoice

Based on the available evidence, SenseVoice is worth considering first if your priority is low-latency, locally deployable speech-to-text; however, the adoption case is stronger for speed and offline use than for being a fully featured speech product. Popularity proof mainly comes from X reposts/discussion and overview posts on Zhihu. Usability proof is better supported by hands-on tests, deployment tutorials, and code-based writeups, which are more useful than ranking-style posts for judging capability and integration effort.

It is not a finished end-user dictation app, and it is not a TTS/voice generation tool like CosyVoice. A more accurate comparison is an open-source ASR/audio-understanding engine that developers embed into products and workflows. The evidence says it supports multilingual ASR plus language identification, emotion recognition, and some event detection such as laughter or applause.