> ## Content Index > Fetch the complete content index at: https://www.thedaringcreatives.com/llms.txt > Use this file to discover other available public pages before exploring further. # Reeps One Learned New Beatbox Phrases From an AI Copy of His Voice - URL: https://www.thedaringcreatives.com/creator-stories/reeps-one-harry-yeff-ai-vocal-twin/ - Published: 2026-11-11T14:00:00.000Z - Updated: 2026-11-11T13:59:59.000Z - Description: Harry Yeff (Reeps One, @reepsone on Instagram) won the UK beatbox title twice. Then he trained an AI on his own voice. It made beatbox phrases he had never made, and he practiced until he could perform against them live. - Author: William Smith - Tags: Creator Stories, artist profile, music, beatbox, voice, ai music, voice model, machine learning, samplernn, elevenlabs, live performance, #medium:Music, #method:Mixed, #handle:@reepsone, #tool:SampleRNN, #tool:ElevenLabs, #issue-2026-10 Harry Yeff performs as Reeps One. He won the UK beatbox championship twice. Then he trained an AI on recordings of his own voice. It came back with beatbox phrases he had never made. "At first the phrases were so complex I could not keep up," he said in [his TEDxBermuda talk](https://www.youtube.com/watch?v=wTMMopLYJn4&ref=thedaringcreatives.com). So he practiced against them until he could perform the duet on stage. He calls the project Second Self. He treated the AI copy of himself as a sparring partner. By his account, it left him able to do things with his voice he couldn't do before. These days he goes by Harry Yeff and posts as [@reepsone on Instagram](https://www.instagram.com/reepsone/?ref=thedaringcreatives.com). His older beatbox videos are on [the Reeps One YouTube channel](https://www.youtube.com/@REEPSONE?ref=thedaringcreatives.com), and the AI work is on [his Vimeo page](https://vimeo.com/reepsone?ref=thedaringcreatives.com). ## How he got into beatboxing Before beatboxing, he played other instruments. "I played trumpet, violin, and drums when I was young, but I was frustrated not having an instrument with me all the time," he told [the Harvard Gazette](https://news.harvard.edu/gazette/story/2016/10/a-beat-all-his-own?ref=thedaringcreatives.com) in 2016\. His voice became the instrument he always had. "My rehearsal studio is with me 24/7," he said. On [the Creativity Squared podcast](https://www.youtube.com/watch?v=u5r-4cg9MMw&ref=thedaringcreatives.com) in 2023, he said he went on to judge the World Championships. He described his range as hundreds of sounds. Reeps One — "The Beginning" (2010), his first upload, on YouTube ([@REEPSONE](https://www.youtube.com/@REEPSONE?ref=thedaringcreatives.com)) ## Why he wanted an AI opponent As a kid he played chess against an AI chessboard. He told the podcast's host, Helen Todd, that the lesson stuck: "In chess my greatest innovations came from my opponents." Second Self puts the AI in the opponent's seat. ## Training the AI on his own voice He built it during art-and-technology residencies. At Harvard he met CJ Carr, who makes AI-generated music as part of Dadabots. Yeff told Todd he wanted to know if machine learning could help him "learn and do things that I couldn't have done without said technology." He trained the model on his own voice, the same starting point [Holly Herndon used for Holly+](https://www.thedaringcreatives.com/creator-stories/holly-herndon-voice-model-holly-plus/). He started with recordings of his speech. Then he added his beatbox sounds. "So the generative audio started to perform, it started to do phrases that I have never done," he said at TEDx. He liked what he heard. Almost no one around him did. "Everybody else found it borderline disturbing," he told Todd. Harry Yeff — ["Second Self: Human and Machine"](https://vimeo.com/827025036?ref=thedaringcreatives.com), filmed in an anechoic chamber, on Vimeo ([reepsone](https://vimeo.com/reepsone?ref=thedaringcreatives.com)) ## Practicing against the AI The model produced hundreds of hours of new phrases in his voice. He went through them and learned them. "I could play phrase chess with myself," he said at TEDx. He told Todd why it worked. After nearly 20 years, his technique had turned into habits. Certain sounds always led to the same next sounds. The model took the sounds he already made and put them together in new orders. He called it "seeing a ghost three years in front of you that you can chase." He told her it improved his speed and control. [His Vimeo upload of the piece](https://vimeo.com/827025036?ref=thedaringcreatives.com) calls the AI tools "a sparring partner or workout buddy." He says it was recorded around 2018, in a soundproof, echo-free room at Bell Labs. ## Performing with it live He also performs Second Self on stage. He trades phrases with the AI voice in front of an audience. His upload of [the TIME100 performance](https://vimeo.com/1052939573?ref=thedaringcreatives.com) says the model produces "new vocal phrases at speeds that test Harry Yeff's world-class abilities." Harry Yeff — "Second Self" live at TIME100, on Vimeo ([reepsone](https://vimeo.com/reepsone?ref=thedaringcreatives.com)) ## Voice agents with ElevenLabs In December 2024, the AI voice company ElevenLabs [named him its Artist-in-Research](https://elevenlabs.io/blog/harry-yeff-joins?ref=thedaringcreatives.com). "I believe that conversational AI can be a design and artistic medium," he said in the announcement. The main project since then is a set of AI voices you can talk to, built for places and animals. "So I started a project to give nature a voice," he said in [a December 2025 Q&A with Drexel](https://drexel.edu/westphal/news-events/news/2025/December/Everyone-Deserves-Creativity-Harry-Yeff/?ref=thedaringcreatives.com). The first is Thwaites Glacier in Antarctica. He told Drexel the glacier's voice can tell stories built on data like 100 years of meltwater change, and guide meditations. It also changes how it talks, from technical to poetic, depending on who's asking. Harry Yeff & ElevenLabs — ["Agentic Thwaites Glacier"](https://vimeo.com/1147719589?ref=thedaringcreatives.com) (2025), on Vimeo ([reepsone](https://vimeo.com/reepsone?ref=thedaringcreatives.com)) Tools he credits himself SampleRNN (Second Self, with CJ Carr of Dadabots) · [ElevenLabs](https://elevenlabs.io/?ref=thedaringcreatives.com) (the voice agents) SampleRNN from his TEDxBermuda and WIRED Impact talks; ElevenLabs from his Drexel Q&A and the ElevenLabs announcement. He told Drexel he's now "designing voices that can tell stories, guide meditations, and unpack complex ideas in articulate and poetic ways."