How Masters of Scale cloned Reid Hoffman's voice using AI

Executive overview

Recording voiceover is slow, error-prone, and hard to schedule — especially for a host who travels constantly. Masters of Scale solved this by partnering with Respeacher, a Ukrainian company, to create a full synthetic clone of Reid Hoffman's voice. The clone is now indistinguishable from the real thing and is being integrated into production.

The show frames the clone as a pure production tool — scripts are still written by humans, approved by Reid, and editorially vetted. The hard questions around disclosure, authenticity, and misuse are unsettled but actively wrestled with.

A synthetic voice removes the logistical bottleneck without displacing the intellectual content.

Why voiceover was a problem

• Reid travels frequently; he needs a quiet space and the right mindset every time he records
• Vocal quality is inconsistent — colds, stumbles, and fatigue all affect output
• Last-minute script changes could require waking a host at 2am for three corrected words
• Scheduling and preparation for any single recording session is disproportionately costly

How Respeacher builds a voice clone

• Respeacher is not text-to-speech; it uses human voiceover artists to perform lines
• The artist's audio is then filtered through an AI model trained on the target voice
• Early drafts of Reid's clone sounded stiff; later iterations are described as indistinguishable from the real thing
• The first production use was Reid's 2023 audiobook Impromptu, saving dozens of recording hours
• Model quality improves rapidly — Respeacher now wants to redo the audiobook on current models

The production team test drive

• Three producers recorded themselves saying Reid's lines, then ran the audio through Respeacher's speech-to-speech conversion
• Conversion took roughly five minutes per file
• Results varied: one output was described as "Reid with a British accent," another surfaced the performer's vocal anxiety
• Adjusting the pitch slider (−6) on a flat female voice produced the most convincing result
• The exercise prompted immediate reflection on how easy it is to fabricate any voice, nationality, or gender

Guardrails and editorial integrity

• Scripts are written by writers, researched by researchers, vetted by Reid, and signed off editorially before any synthetic voice is used
• The team draws a hard line: Respeacher is a production tool, not a content generator
• Staff writer's framing: "We're not handing over the keys to the car — we're changing the paint job"
• Reid's enthusiastic consent was cited as the key ethical anchor for the team

Disclosure and the listener's bill of rights

• The show has not yet settled on a specific disclosure format — episode audio, show notes, or embedded metadata are all under consideration
• Reid's position: signal AI use clearly, but no line-by-line accounting is required — similar to "names changed to protect the innocent" disclaimers
• The analogy to CGI in films: studios don't list every VFX shot, because it's understood as part of production
• The distinction that matters: did the named person own and approve the content? If yes, attribution holds
• Masters of Scale published a proposed industry standard at mastersofscale.com/pledge

Reid's broader vision for digital twins

• Reid also built a generative video version of himself called Reid AI, trained on 20 years of his content
• Intended use cases include producing content in multiple languages without Reid learning them
• The goal is more output for the audience, not less effort from the team
• Reid's framing: AI tools are like phones and computers — instrumental, not identity-replacing