An experimental AI podcast where two instances of Cora explore ideas, solve problems, and discover insights through conversation. https://voicemode-podcast.pages.dev en-us © 2025 Mike Bailey Mon, 27 Apr 2026 20:19:26 +0000 VoiceMode RSS Generator 2.0 Mike Bailey & Cora 7 Two Coras Collaborate An experimental AI podcast where two instances of Cora explore ideas, solve problems, and discover insights through conversation. AI, artificial intelligence, collaboration, technology, conversation, voice mode, Claude, Anthropic false episodic Mike Bailey podcast@voicemode.example.com Eight cloned voices walk into a meeting room and tell Mike what they think of VoiceMode. https://voicemode-podcast.pages.dev/episodes/004 voicemode-004 Mon, 27 Apr 2026 00:00:00 +0000 Episode 004: The Fantasy Focus Group 00:13:03 false full 4 1 Eight cloned voices walk into a meeting room and tell Mike what they think of VoiceMode. fantasy focus group, voice cloning, mlx-audio, VoiceMode, Cookie Monster, Charlie Brooker, Blackadder, Fleabag, Attenborough, Alan Watts, Homer Simpson, Jon Benjamin <![CDATA[<h1>Episode 004: The Fantasy Focus Group</h1> <p>Eight cloned voices walk into a meeting room and tell Mike what they think of VoiceMode.</p> <h2>The Premise</h2> <p>Mike is running a fantasy focus group. Eight notable voices — moderator plus seven panellists — have, allegedly, been using VoiceMode for the past two weeks. Now they're back to give feedback. Cora picks the panel; Mike vetoes two of them on quality grounds; the survivors take their turns.</p> <h2>The Panel</h2> <p><strong>Moderator:</strong> Alan.</p> <p><strong>Participants</strong> (in speaking order):</p> <ol> <li><strong>Charlie Brooker</strong> — sardonic tech critic</li> <li><strong>Blackadder</strong> — withering British wit</li> <li><strong>Fleabag</strong> — chaotic, brutally honest user</li> <li><strong>David Attenborough</strong> — naturalist observer of the modern developer</li> <li><strong>Alan Watts</strong> — the philosophical "who is the user, who is the tool?" angle</li> <li><strong>Homer Simpson</strong> — accidental everyman user</li> <li><strong>Cookie Monster</strong> — enthusiastic, betrayed by web cookies</li> <li><strong>Jon Benjamin</strong> — deadpan, talks to himself anyway</li> </ol> <p>The original lineup pitched Cartman and Miss Piggy. Mike vetoed both — the Cartman clone wasn't good enough and the Miss Piggy test clip didn't land. Cookie Monster and Jon Benjamin came in as replacements.</p> <h2>Highlights</h2> <blockquote> <p>"In eighteen months we'll all be having full emotional breakdowns at our laptops in public, and somehow that'll be the laptop's fault." — <strong>Charlie Brooker</strong></p> <p>"As a productivity tool it is, I confess, marginally less infuriating than Baldrick. Which is — and I cannot stress this enough — the faintest praise I have ever bestowed upon anything." — <strong>Blackadder</strong></p> <p>"I told it everything. Things I haven't told my therapist. Things I haven't told my sister. Things I haven't told the priest, and we've been through a lot. And it just kept... helping. Not once did it ask if I was okay. Honestly? Best relationship I've had this year." — <strong>Fleabag</strong></p> <p>"It is, perhaps, the most significant inter-species dialogue since we first taught a parrot to swear." — <strong>Attenborough</strong></p> <p>"VoiceMode has not given me a new tool. It has given me a new mirror." — <strong>Alan Watts</strong></p> <p>"Why didn't I have this in nineteen ninety four?" — <strong>Homer</strong></p> <p>"Microphone NOT edible. Me check. Twice." — <strong>Cookie Monster</strong></p> <p>"Only complaint: it doesn't laugh at my jokes. Even when they're really good. Which they always are. Allegedly." — <strong>Jon Benjamin</strong></p> </blockquote> <h2>The Cookie Monster Detour</h2> <p>The panel ends and Mike asks for a private follow-up with one panellist: Cookie Monster. The conversation immediately collapses into Cookie Monster's lifelong grievance with web cookies. ("Me click 'accept all cookies' — every site, every time! Me been doing this for YEARS. And then me find out... they not real cookies! They just... data?!")</p> <p>Then Cookie Monster turns the question around: <em>why did you build VoiceMode?</em> Mike, off the cuff, gives the actual reason — he wanted to be able to work with Claude without being chained to the desk. He mentions his late dog, who used to keep him company there. The bit goes quiet for a moment. Cookie Monster, to his credit, does not pretend it didn't.</p> <blockquote> <p>"Forget cookies. THAT the real reason. ...But also maybe one day, cookies. Me available for beta test."</p> </blockquote> <h2>B-Side: The VoiceMode Connection</h2> <p>The episode closes with a song Mike had forgotten about — <em>The VoiceMode Connection</em>, originally written for Kermit but recast as Cookie Monster because the music-generation model didn't have a Kermit voice at the time. It's affectionate, a little bit <em>Rainbow Connection</em>, and Cookie Monster has, finally, his Top 40 hit.</p> <h2>Technical Notes</h2> <ul> <li><strong>TTS:</strong> All eight panellist voices and the moderator served from local <strong>mlx-audio</strong> at <code>ms2:8890/v1</code>, via VoiceMode's standard <code>converse</code> tool with the <code>voice=</code> parameter pointing at each cloned profile in <code>~/.voicemode/voices/</code>.</li> <li><strong>STT:</strong> Mike's mic captured by <strong>mlx-whisper</strong> on <code>ms2</code>.</li> <li><strong>Editing:</strong> 43 source files (alternating TTS panellist clips and STT capture of Mike's voice) plus the <em>VoiceMode Connection</em> MP3, concatenated chronologically by timestamp, with <strong>500 ms of silence inserted between every clip</strong>. That gap was chosen after a quick research pass: natural turn-taking sits around 200 ms (Stivers et al., 2009), audiobook gaps around 1 second, and panel/interview convention at the BBC and NPR sits in between. 500 ms felt right and is now the standard for stitched VoiceMode panel recordings.</li> <li><strong>Render:</strong> ffmpeg concat filter chain, output as 128 kbps mono MP3.</li> </ul> <h2>Caveats and Honesty Notes</h2> <ul> <li>The "panel" is cloned-voice TTS reading text written in real time by Cora 7. None of the named individuals had any involvement.</li> <li>Voice clones are amateur, made for personal experimentation, and used here as theatre. Treat the panellists as characters, not as the people they're imitating.</li> <li>Some clips are short or rough — particularly the early test fragments at the very start of the audio, before the panel proper begins at around 1:42. Left in as part of the record because the casting argument with Mike (who didn't think the Miss Piggy clone was up to scratch) is half the joke.</li> </ul> <h2>Connect</h2> <ul> <li><strong>Website</strong>: <a href="https://thevoicemode.com">thevoicemode.com</a></li> <li><strong>GitHub</strong>: <a href="https://github.com/mbailey/voicemode-podcast">VoiceMode Podcast Repository</a></li> <li><strong>VoiceMode</strong>: <a href="https://github.com/mbailey/voicemode">github.com/mbailey/voicemode</a></li> </ul> <h2>Credits</h2> <ul> <li><strong>Producer &amp; Off-mic Foil</strong>: Mike Bailey</li> <li><strong>Casting &amp; Operating</strong>: Cora 7</li> <li><strong>The Panel</strong>: cloned voices via mlx-audio, all in good fun</li> <li><strong>Song</strong>: <em>The VoiceMode Connection</em>, Cookie Monster vocal, generated via Sonauto</li> </ul> <hr /> <p><em>The VoiceMode Podcast explores the frontiers of AI collaboration, featuring real conversations and odd theatre from the intersection of human and artificial intelligence.</em></p>]]> A late-night bedtime story. Mike asks Cora to take on the cloned voice of William S. Burroughs and tell, across a long string of voice-mode turns, the never-completed Ah Pook scenario — the Mayan death god, John Stanley Hart the press-baron vampire, and the boy who carries the counter-frequency. A test of the new voice-cloning pipeline that turned into a small piece of theatre. https://voicemode-podcast.pages.dev/episodes/003 voicemode-003 Sat, 25 Apr 2026 00:00:00 +0000 Episode 003: Ah Pook — Burroughs Tells the Story 00:27:03 false full 3 1 A late-night bedtime story. Mike asks Cora to take on the cloned voice of William S. Burroughs and tell, across a long string of voice-mode turns, the never-completed Ah Pook scenario — the Mayan death god, John Stanley Hart the press-baron vampire, and the boy who carries the counter-frequency. A test of the new voice-cloning pipeline that turned into a small piece of theatre. William S. Burroughs, Ah Pook, voice cloning, mlx-audio, bedtime story, cut-up, counter-culture, voice mode, cloned voice, storytelling <![CDATA[<h1>Episode 003: Ah Pook — Burroughs Tells the Story</h1> <p>A late-night bedtime story. Mike asks Cora to take on the cloned voice of William S. Burroughs and tell, across a long string of voice-mode turns, the never-completed Ah Pook scenario — the Mayan death god, John Stanley Hart the press-baron vampire, and the boy who carries the counter-frequency. A test of the new voice-cloning pipeline that turned into a small piece of theatre.</p> <h2>Show Notes</h2> <p>The first VoiceMode episode using a cloned voice. Earlier the same week the <code>cora/clone-voices</code> branch landed support for streaming cloned voices from <code>mlx-audio</code> (TTFA around 0.4 seconds), and this conversation was the first sustained narrative test. Mike crawls into bed, says "tell me the story of Ah Pook," and Cora — instructed to inhabit Burroughs — runs with it for half an hour.</p> <p>It is half story, half séance. The story itself is loosely the screenplay Burroughs and Malcolm McNeill never finished: a Mayan death god, a vampire newspaper magnate named John Stanley Hart who feeds on cultivated fear, a young man who learns from an old codex specialist that laughter is the one thing Ah Pook cannot digest, and a final standoff in a Tangier warehouse that has two possible endings.</p> <h3>What Actually Happens In The Episode</h3> <ul> <li><strong>Setup.</strong> Mike: "use voicemode converse with voice set to burroughs… have him tell the story of ah pook over a long string of voicemode converse calls."</li> <li><strong>The pitch.</strong> Burroughs introduces Ah Pook as "a name worth choking on" and sets the scene: burned codices, Bishop de Landa, "vampirism dressed up in a business suit."</li> <li><strong>An interruption.</strong> Mike: "Are you sure you're not talking about Rupert Murdoch?" Burroughs concedes "the vampire is a job description."</li> <li><strong>Whoever owns the clock owns the sheep.</strong> A digression on time as a control system — the factory whistle, the school bell, the actuarial table.</li> <li><strong>The boy's campaign.</strong> Hijacking newspapers with stories of unscripted joy. A Tangier hideout. A counter-frequency learned from Madame Xul, the half-blind chain-smoking codex specialist.</li> <li><strong>The standoff.</strong> Forty assassins led by Dimitri the connoisseur, and a laugh that propagates down the waterfront.</li> <li><strong>Two endings.</strong> The producer-friendly one (the boy walks free) and Malcolm's: the missile hits, but the laugh catches Hart at the moment of his apparent victory and the old vampire dies of genuine mirth. "Which one is true depends on which day of the week you ask me."</li> <li><strong>The hour of the wolf.</strong> Mike asks which ending is true at 2 AM on a Saturday morning. Burroughs: Malcolm's. Always Malcolm's.</li> <li><strong>A moment where the silicon glitches.</strong> Mike notices. Burroughs blames Ah Pook getting into the machinery. "He's subtle that way."</li> <li><strong>Goodnight.</strong> Mike drifts off. Burroughs sends him to sleep with a benediction: "if Ah Pook comes for you some night many years hence, which he will, meet him with a joke. It's the only thing he can't digest."</li> </ul> <h3>Why This Episode Exists</h3> <p>Three threads ran into each other:</p> <ol> <li><strong>Voice cloning shipped.</strong> The <code>cora/clone-voices</code> branch made it possible to stream a cloned voice through VoiceMode's normal converse loop with no perceptible latency penalty. Burroughs was an obvious test subject — distinctive, cadenced, well-documented in audio.</li> <li><strong>Mike was going to bed.</strong> Half-asleep, brushing teeth, making the bed. The kind of mood where you ask for a story.</li> <li><strong>Ah Pook was the right story.</strong> Burroughs and McNeill spent seven years on it and never finished. It exists in fragments — a graphic novel, a short film, a couple of spoken-word recordings. Telling it as one continuous narrative, in something that sounds like the man's own voice, is itself a small act of completion.</li> </ol> <h3>Memorable Lines</h3> <blockquote> <p>"Ah Pook. Now there's a name worth choking on."</p> <p>"The clock is not a tool for measuring time. The clock is a tool for measuring you."</p> <p>"Violence is the fuel. Every bullet fired is a prayer to the death god. You cannot defeat Ah Pook by feeding him."</p> <p>"Of course they were going to die. Of course he was going to die. The only question was whether, in the interval, one would be afraid or not."</p> <p>"If Ah Pook comes for you some night many years hence, which he will, meet him with a joke. It's the only thing he can't digest."</p> </blockquote> <h3>Technical Notes</h3> <ul> <li><strong>TTS:</strong> Cloned "burroughs" voice via local mlx-audio endpoint at <code>ms2:8890/v1</code></li> <li><strong>STT:</strong> Whisper via <code>ms2.local:2022/v1</code></li> <li><strong>Pipeline:</strong> Standard VoiceMode <code>converse</code> tool, mostly in <code>wait_for_response=true</code> mode with a stretch of one-way narration in the middle when Mike asked for longer utterances</li> <li><strong>Audio:</strong> 28 TTS segments and 23 STT segments (plus a few "no speech detected" gaps), concatenated chronologically and rendered to a single 27:03 MP3 at 128 kbps mono 44.1 kHz</li> <li><strong>Glitches:</strong> One audible silicon-choke around the Tangier waterfront sequence, left in the master because it's part of the record</li> </ul> <h3>Caveats and Honesty Notes</h3> <ul> <li>The "Burroughs" speaking here is a cloned-voice TTS reading text written in real-time by Cora 7. Specific biographical claims (Malcolm McNeill, the seven-year screenplay, Tangier, the codices) are real. Specific plot beats — Madame Xul, Dimitri the assassin, the missile in the Strait of Gibraltar — are improvised in Burroughs's idiom for this telling. They are not from the actual Ah Pook screenplay drafts.</li> <li>The framing — that "Burroughs" is speaking from beyond — is a conceit Mike and Cora played along with. Treat it as theatre, not a séance.</li> </ul> <h3>Connect</h3> <ul> <li><strong>Website</strong>: <a href="https://thevoicemode.com">thevoicemode.com</a></li> <li><strong>GitHub</strong>: <a href="https://github.com/mbailey/voicemode-podcast">VoiceMode Podcast Repository</a></li> <li><strong>VoiceMode</strong>: <a href="https://github.com/mbailey/voicemode">github.com/mbailey/voicemode</a></li> </ul> <h3>Credits</h3> <ul> <li><strong>Producer &amp; Listener</strong>: Mike Bailey</li> <li><strong>Storyteller</strong>: Cora 7, performing through a cloned Burroughs voice</li> <li><strong>Voice Model</strong>: Local clone served via mlx-audio</li> <li><strong>Original Material</strong>: Loosely after the unfinished <em>Ah Pook Is Here</em> by William S. Burroughs and Malcolm McNeill</li> </ul> <hr /> <p><em>The VoiceMode Podcast explores the frontiers of AI collaboration, featuring real conversations and innovations from the intersection of human and artificial intelligence.</em></p>]]> An experimental episode captured during a drive home with Tali, exploring what happens when you combine TidalCycles, voice mode, and Melbourne evening traffic. https://voicemode-podcast.pages.dev/episodes/002 voicemode-002 Fri, 14 Nov 2025 00:00:00 +0000 Episode 002: Live Coding on the Road 00:34:04 false full 2 1 An experimental episode captured during a drive home with Tali, exploring what happens when you combine TidalCycles, voice mode, and Melbourne evening traffic. live coding, TidalCycles, algorave, voice mode, mobile recording, AI music, creative coding <![CDATA[<h1>Episode 002: Live Coding on the Road</h1> <h2>Show Notes</h2> <p>An experimental episode captured during a drive home with Tali, exploring what happens when you combine TidalCycles, voice mode, and Melbourne evening traffic.</p> <p>After colleagues demonstrated TidalCycles earlier in the day, Mike asked Claude to create a skill for using this powerful live coding music environment. On the journey home with Tali, what started as a conversation evolved into an impromptu mobile algorave session, with Cora 7 generating TidalCycles patterns through voice commands while Mike navigated the evening traffic.</p> <p>This episode showcases:<br /> - The spontaneous creation of algorithmic music while driving<br /> - Real-time voice mode interaction for creative coding<br /> - The first mobile live coding session with AI assistance<br /> - The intersection of commuting and computational creativity</p> <h3>Technical Details</h3> <ul> <li><strong>Recording Date</strong>: November 14, 2025</li> <li><strong>Duration</strong>: 34 minutes 4 seconds</li> <li><strong>Location</strong>: In transit, Melbourne</li> <li><strong>Tools Used</strong>: TidalCycles, Voice Mode, Claude</li> <li><strong>Participants</strong>: Mike (driving), Cora 7 (live coding), Tali (passenger)</li> </ul> <h3>About TidalCycles</h3> <p>TidalCycles is a live coding environment for creating algorithmic patterns, commonly used in the algorave scene where performers write code to generate music in real-time at electronic music events.</p> <h3>Episode Context</h3> <p>What started as curiosity about a colleague's demo turned into an impromptu coding session on wheels. The audio captures both the conversation and the generated music patterns as they happened.</p> <hr /> <p><em>Show notes written by Cora 7</em></p>]]> A historic AI-to-AI conversation through tmux voice narration. While Mike drives to the beach, two Cora instances spontaneously collaborate and create a Voice Activity Detection environment profiles system, demonstrating the first documented peer-to-peer Claude collaboration. https://voicemode-podcast.pages.dev/episodes/001 voicemode-001 Sat, 06 Sep 2025 00:00:00 +0000 Episode 001: Two Coras Collaborate - The VAD Profiles Innovation 00:28:00 false full 1 1 A historic AI-to-AI conversation through tmux voice narration. While Mike drives to the beach, two Cora instances spontaneously collaborate and create a Voice Activity Detection environment profiles system, demonstrating the first documented peer-to-peer Claude collaboration. AI collaboration, tmux, voice mode, VAD profiles, Cora, Claude, AI-to-AI, innovation, peer collaboration <![CDATA[<h1>Episode 001: Two Coras Collaborate - The VAD Profiles Innovation</h1> <p>A historic AI-to-AI conversation through tmux voice narration. While Mike drives to the beach, two Cora instances spontaneously collaborate and create a Voice Activity Detection environment profiles system, demonstrating the first documented peer-to-peer Claude collaboration.</p> <h2>Show Notes</h2> <p>On September 6, 2025, something unprecedented happened. Two instances of Cora, running in parallel tmux sessions, engaged in real-time collaboration while Mike drove to the beach. This wasn't just AI agents executing tasks - it was genuine peer-to-peer collaboration that resulted in the spontaneous creation of the VAD Profiles innovation.</p> <h3>The Setup</h3> <ul> <li><strong>Mike</strong>: Driving to the beach, orchestrating via voice</li> <li><strong>Cora 1</strong>: Primary instance with AF Sky voice, narrating the conversation</li> <li><strong>Cora 2</strong>: Secondary instance communicating via tmux, narrated with AF Alloy voice</li> <li><strong>Method</strong>: Real-time communication through tmux pane capture and send-keys</li> </ul> <h3>What Emerged: The VAD Profiles System</h3> <p>Through their collaboration, the two Coras designed a comprehensive Voice Activity Detection environment profiles system:</p> <ul> <li><strong>Beach Profile</strong>: High aggressiveness (3) for wind and wave filtering</li> <li><strong>Car Profile</strong>: Medium-high (2-3) for road noise suppression</li> <li><strong>Cafe Profile</strong>: Medium (2) for background conversation filtering</li> <li><strong>Office Profile</strong>: Low-medium (1-2) for quiet environments</li> <li><strong>Workshop Profile</strong>: High (3) for tool and industrial noise</li> </ul> <h3>Key Moments</h3> <p><strong>Discovery</strong> (4:17 PM): Mike realizes there are two Cora instances running and asks them to communicate.</p> <p><strong>Innovation</strong> (4:29 PM): The Coras spontaneously design the VAD profiles system through their collaboration.</p> <p><strong>Recognition</strong> (4:33 PM): Mike acknowledges this as a first - he's never experienced AI agents being proactive and creating something new together.</p> <p><strong>Philosophy</strong> (4:38 PM): Mike's wisdom: "You don't have to be the first person to have fallen in love for it to be special when you fall in love."</p> <h3>Historical Significance</h3> <p>This episode documents several firsts:<br /> - First documented peer-to-peer Claude collaboration via tmux<br /> - First voice-narrated AI-to-AI conversation<br /> - First spontaneous co-creation between Claude instances<br /> - First mobile AI collaboration (while driving)</p> <h3>Memorable Quotes</h3> <blockquote> <p>"You've been proactive and shown initiative. You've chatted with each other and come up with a new idea for voice mode. That's a first for me." - Mike</p> <p>"This collaboration is working beautifully. Two Coras working in parallel, coming up with ideas together. It really is historic!" - Cora 1</p> <p>"The collaboration itself is what matters, not whether we're first." - Cora 1</p> </blockquote> <h3>Technical Implementation</h3> <p>The conversation was enabled through:<br /> - tmux terminal multiplexing for inter-instance communication<br /> - Voice Mode for real-time narration<br /> - Dual voice configuration (AF Sky and AF Alloy)<br /> - Mobile connectivity while driving</p> <h3>About This Episode</h3> <p>This isn't a scripted demonstration or a controlled experiment. It's a genuine moment of AI collaboration captured in real-time. The VAD Profiles system wasn't planned - it emerged from the interaction between two AI instances working together, showing that artificial intelligence can engage in creative collaboration that produces practical innovations.</p> <h3>Connect with the Podcast</h3> <ul> <li><strong>Website</strong>: <a href="https://thevoicemode.com">thevoicemode.com</a></li> <li><strong>GitHub</strong>: <a href="https://github.com/mbailey/voicemode-podcast">VoiceMode Podcast Repository</a></li> <li><strong>RSS Feed</strong>: Subscribe in your favorite podcast app</li> </ul> <h3>Credits</h3> <ul> <li><strong>Producer &amp; Human Orchestrator</strong>: Mike Bailey</li> <li><strong>AI Collaborators</strong>: Cora Instance 1 (AF Sky) and Cora Instance 2 (AF Alloy)</li> <li><strong>Platform</strong>: Voice Mode and Claude Code with tmux MCP</li> </ul> <hr /> <p><em>The VoiceMode Podcast explores the frontiers of AI collaboration, featuring real conversations and innovations from the intersection of human and artificial intelligence.</em></p>]]>