PlayAI

The MrBeast onboarding doc, as read by AI MrBeast

Today, we’re excited to announce $21 million in new capital from Kindred Ventures, Race Capital, 500 Global, Y Combinator, Soma Capital, Pioneer Fund and others to build the future of conversation with real-time voice AI. We’re on a mission to make voice AI platforms simple and accessible to all businesses, developers and creators. And with our newest PlayDialog model, conversations sound real, emotive and human-like – how they should be. We’ve come a long way since building our first Chrome extension. Today, our low-latency generative voice models are powerful, accessible, and serve applications as diverse as 24/7 customer support, appointment scheduling and sales lead engagement. We can’t wait to show you what’s possible with voice AI. Read more ⬇️⬇️⬇️

There has been a long history of trying to replicate human voice. Elevators used to have attendants, but they were replaced with a human voice (“Doors closing,” “Going up!”) as a way to provide assurance to the riders. More recently publications deployed “text-to-speech” buttons to allow you to listen to articles. All of these attempts were unbearable. People couldn’t listen to a robotic voice for sustained periods. I think the “uncanny valley” of seeing AI image slop is even worse when listening to AI voice. Until now. I spoke with one of the leading companies that have the industry’s SOTA model for voice agents, PlayAI. (Disclaimer: I led 500 Global’s investment in PlayAI more than 2 years ago). I had a fun conversation with co-founder and CEO Mahmoud Felfel Some of the topics discussed include: • What makes it difficult to achieve human level voice by AI? • What are some recent breakthroughs that unlocked more capability in this area? • What is the future of voice agents and how humans will relate with synthetic voice? Enjoy the discussion! https://lnkd.in/gk5FhntY

Our new voice model PlayDialog beta is launching on Product Hunt today! It's the world's first voice AI model built for multi-turn conversations & uses the entire conversation history to generate speech with expressive tone, prosody, and emotion. Check it out: https://lnkd.in/gGgvGxSh

This is a fun one -- our new model is crazy expressive, and play.ai/playnote is an awesome interface to try it out!

Wake up friend, there's a new usecase for playnote -- 🎨 to 📕 (image to story). We used Black Forest Labs through fal to create an image from a few words and then fed that into playnote. Instant storybook for your kids. You can even clone your own voice to read to them! And the whole thing took <5 clicks.

Today, we’re introducing PlayDialog. It’s a SOTA voice model built for fluid, emotive, human-like conversation. PlayDialog lets you create engaging dialog and narrative from one or two speakers programmatically, and from documents, website links, or text. It takes the full history of previous prompts and responses as input to generate a response that’s emotionally coherent and fluid. Nothing else sounds as human. But... Hearing is believing. So alongside PlayDialog, we’re launching PlayNote, too. Just drop in any files, website links, and get beautifully narrated podcasts, briefings, stories, and more. And the best part? The whole thing is accessible via PlayNote API. Full announcement linked in the comments.

Releasing our new API that lets you embed agents right onto your website with just a few lines of code! BYO LLM, clone whatever voice you want, define whatever functionality you need! This is a small step towards voice as the new computer interface. Docs: https://lnkd.in/gajZGb8N Try our demo out yourself: https://lnkd.in/gNU9RXVj

Did you know that we have Zapier and Make integrations with our agents? That means you can trigger actions to over 7000 apps from our agents! Sending emails, updating your CRM, querying order status, creating Zendesk tickets -- your agent can handle it all for you!

Today we’re introducing our latest Text-To-Speech model, Play 3.0 mini.  It’s faster, more accurate, handles multiple languages, supports streaming from LLMs, and it’s more cost-efficient than ever before. 🚄 It’s faster.  We now support websockets, and using those, our SDK gets time to first audio (TTFA) of 189ms, making it the industry’s fastest model.  It also supports text-in streaming from LLMs. 🎯 It’s more accurate.  We trained it on a diverse dataset of alpha-numeric phrases so it accurately handles phone numbers, passport numbers, dates, currencies, etc. and it has a character error rate 40% lower than Play 2.0. 🧮 It now reads numbers and acronyms just like humans do.  Numbers like phone numbers, and acronyms are identified by the model, and read out with more natural pacing, making for a more natural conversation. 💰 It’s more cost efficient.  So, we’re lowering the cost per thousand characters for our higher volume tiers, and launching a competitive “Pro” tier at $49 a month for users with more modest volume requirements. Check it out for yourself! Playground: https://lnkd.in/gjdzvUCb Blog post: https://lnkd.in/gZRntMqd