For meeting customers, the AI interpreter's voice for a language could stop for the rest of a meeting once it fell a few seconds behind. Now every language keeps its voice: when the interpreter falls behind, it speaks a little faster, and if it is still far behind, it skips sentences it has not started, never one it is saying, while the captions keep every sentence. The first words of a sentence are no longer cut off. The speaker's own voice plays softly under the translation and at your level between sentences, and you can turn it off. A speaker's tile shows the languages their words are being translated into, and the interpreter listens only to people's microphones, never to its own voice.