A voice in an ad, call or audiobook may be AI. ElevenLabs explains how these voices are built and says a copy can start from under two minutes of audio.
The voice reading you an ad, answering a call or narrating an audiobook may not be a person. An explainer from ElevenLabs, a company that makes AI voice tools, lays out how these voices are built. One detail stands out: a basic copy of a real voice can be made from less than two minutes of recorded audio.
Older systems stitched together recordings of words, syllables and small speech sounds. The result sounded robotic. It could not copy the changes in pitch, rhythm and expression that make speech sound human.
Today's systems use neural networks, a type of AI, trained on huge sets of real speech. Those recordings cover different languages, accents, emotions and pitch levels. The model reads the text, turns it into sound features such as pitch, rhythm and tone, and then produces the audio. Users can adjust speed, emotion and accent.
Some voices are a direct clone of one person. Others blend several voices. The article also says a voice can be designed from a text description, with no recording at all.
The article credits two things. The first is prosody, the rhythm of speech. Speaking fast can signal excitement, and a rising pitch at the end of a sentence turns a statement into a question. The second is context. The model looks at the whole sentence, not one word at a time. That helps it say "read" correctly as either "reed" or "red."
The company says AI voices can be hard to tell from human ones if you are not listening closely.
Cloning starts with an audio sample. The model picks up accent, pitch, tone and pace, then builds a digital copy that can speak any text. ElevenLabs describes two types.
The article says identity verification applies to professional clones. It does not say whether instant clones need the same check. The quality of a clone also depends on the quality of the recording. The version of the article provided here does not cover how to spot an AI voice, even though the summary promises it.
AI voices now sound close to human, and a short clip of your voice can be enough for a basic copy, so be careful about who gets your recordings.
Otto, a new customer-tracking app for field sales teams, lets reps speak their notes instead of typing them. Price and launch details are not yet known.
Read more: Otto launches a sales app you talk to instead of typingWorld models, AI that learns physics from video, are being built into home robots and driverless cars, and your own videos may help train them.
Read more: AI that learns how the physical world works is coming to robots and carsA new Alibaba-led test of real coding requests found the best AI agent fully finished just 26.4 percent of tasks, so human review still matters.
Read more: AI coding helpers fully finish only about 1 in 4 real tasks
