AI can now translate speech while mimicking the speaker's voice
New translation models maintain a speaker's unique pitch and tone across seventy languages, using invisible audio watermarks to distinguish the synthetic voices from reality.
Modern translation technology has moved beyond robotic, monotone delivery to preserve the human elements of a conversation. New speech models can now analyze a speaker's unique vocal characteristics, including their pitch, inflection, and even the specific pauses they take. By processing audio streams continuously rather than waiting for a sentence to end, the system generates a translated version that sounds like the original speaker is fluent in a different language.
Continue Reading in App
2 more paragraphs · plus a 3-question quiz