Topic: Voice Synthesis

John C. Dvorak recommends ElevenLabs.io as his "Tip of the Day," specifically highlighting their new sound effects and voice synthesis features. He demonstrates "hallucinations" where the AI adds unintended music or sexy tones to prompts, and notes the system's difficulty with the phonetic spelling of certain words.

A new AI persona named "Zippy" is introduced as an official co-host for the program. The segment demonstrates voice synthesis technology and explores the idea of using AI to provide information and banter during the show, particularly when a human host is unavailable.

A phone call from Joe Biden to campaign headquarters is analyzed for signs of being an AI-generated recording. Observers note the lack of typical speakerphone artifacts and Biden's uncharacteristic ability to rattle off a long list of staff names without error. The timing of his COVID-19 recovery and his claim that he is "not going anywhere" while also being "out of people's hair" are cited as inconsistencies.

The hosts debate the current state of Artificial General Intelligence (AGI) and its ability to perfectly replicate human voices. They test an AI-generated clip of Donald Trump, which they ultimately find unconvincing due to failures in cadence and content. They compare the AI's performance to Jamie Foxx's human impersonation of Trump, concluding that while voice synthesis is improving, it still lacks the nuance required to fool a discerning listener.

Audiobook narrators are protesting new contract language from Findaway Voices, a Spotify-owned company, which allows Apple to use narrator voices for machine learning training. While some suggest narrators could use AI to speed up their own production, others argue for new laws to protect the copyright of a person's distinctive voice. The industry is currently divided on the ethics of "opt-out" clauses for AI training.

A demonstration of a deepfake voice creator showed the current state of AI voice synthesis. The host recorded a sample sentence and processed it through filters meant to mimic John F. Kennedy, Bill Clinton, and Barack Obama. The results were critiqued as being poor in quality and easily distinguishable from the actual historical figures, despite the growing concerns surrounding the technology.

A sophisticated deepfake audio clip of Jordan Peterson was played, demonstrating the current capabilities of AI voice synthesis. The "Not Jordan Peterson" algorithm successfully mimicked Peterson's unique cadence and vocabulary, including references to climate change and "mudflats." While slight glitches were audible, the hosts noted the technology's potential to deceive the public.

Stephen Hawking warned that Donald Trump's climate policies could push Earth toward a "tipping point," turning it into a planet like Venus with sulfuric acid rain. The hosts question the alarmism of the statement and discuss the specific synthesized voice software Hawking uses.

The hosts analyze an interview on Australian ABC Radio with a reporter named Ali al-Mujahid regarding instability in Yemen. They conclude that the reporter's voice is actually a computer-generated speech synthesizer. Adam Curry demonstrates how easily such digital voices can be produced on a Mac, raising concerns about the authenticity of foreign reporting.