Tencent open-sources AI interpreter Hy-SimulMT that responds in 3 seconds

Tencent has released and open-sourced its simultaneous interpretation model Hy-SimulMT, which translates speech between Chinese and English in real time. The model listens and translates at the same time, outputting text and voice instantly. It can even clone the speaker's voice and tone on the fly. It is now live on Tencent's Hy translation app for free. Low latency is its key strength, responding in about 3 seconds, faster than the 4 to 5 seconds human interpreters usually need. On the open-source benchmark RealSI for long audio, it beat mainstream open-source models and most commercial APIs in both speed and quality. To test it fairly, the team gathered speech, interview, and meeting audio across ten fields such as software, medicine, law, and education, and compared it with 6 certified human interpreters. The result: Hy-SimulMT matched their translation quality while responding even faster.