The AI Voice Tool You're Already Paying For Just Got a Lot More Powerful

ElevenLabs packed a lot into September. If you use ElevenLabs for voice, transcription, or music generation in your business, the September developer update is worth ten minutes of your time. The headline is Eleven v4, their new text-to-speech model. It's now ranked first by Artificial Analysis, and in blind listening tests roughly three out of four people preferred it over what came before. The difference isn't just clearer audio. The model actually reads context: tone, pacing, emotion, across more than 90 languages. Think of it as the difference between a GPS that reads you an address and one that notices you're driving in a school zone and adjusts accordingly.

There's also a Turbo version that responds in about 100 milliseconds. That's fast enough for real-time voice agents where a half-second delay breaks the illusion. If you're running a phone bot for booking, intake, or support, that latency gap is the whole game.

The update also includes a new Image and Video API. Eighteen models, including Veo 3.1 and GPT Image 2.5, all accessible with the same API key you already use for speech. One key, one account, images and video. That's genuinely useful for small shops that don't want to manage five different vendor relationships. And a feature called Structured Procedures (still in alpha) lets voice agents run a checklist of steps in the same order every single time. Useful if you've ever had a voice agent go rogue mid-call and skip a required question. The Scribe v2 Medical transcription model is also now generally available, cutting word errors on clinical audio by 35%. Relevant if you're anywhere near healthcare. And Music v2.5 is live on the API, preferred in blind tests across nearly 48,000 prompt pairs, with the biggest lead on vocal-led tracks.

If you're building on Claude, check your model version today. Anthropic just issued a deprecation notice for Claude Sonnet 4.5. The model gets retired on November 24, 2026. That sounds far off, but starting October 30, anyone still calling the Sonnet 4.5 API could start seeing rate errors and reduced availability when traffic spikes. The replacement is Claude Sonnet 5, which Anthropic says delivers better results at a lower price. If you have a developer or agency managing your AI integrations, forward them this and ask them to confirm the migration is on the calendar. If you're in Claude Code yourself, there's a built-in migration command that updates model names and settings automatically. Don't wait until November.

Plaud's new Agent feature turns recordings into actual work output. Plaud is the little recorder device a lot of owners carry to capture meetings and calls. This October they're rolling out Plaud Agent, which adds a credits-based layer on top of your existing subscription. The core recording and transcription stays as-is. The new stuff: you can ask questions across any recording or your entire archive, trigger real actions like sending an email or creating a Notion page directly from what you just said, and export formatted documents (Word, PowerPoint, PDF) from a recording. There's also a routines feature for scheduled or saved instructions. Think of it as the difference between a note-taking app and a junior assistant who actually does something with the notes. Welcome credits are included for the first three months based on your plan tier.

-Scott

SmartOwner is published (almost) daily by the team at DigitalTreehouse. Want AI consulting or automations for your business? Reply to this email.