Podonos

We are thrilled to announce that Podonos has raised $2.4M in pre-seed funding 🎉 This milestone helps us tackle one of the most overlooked challenges in voice AI: evaluation. Voice AI is moving fast — from text-to-speech and voice cloning to multilingual assistants. But without accurate and scalable evaluation, teams risk missing the unknown issues that matter most in the real world. That is exactly what Podonos was built to solve. Our human-in-the-loop platform delivers evaluation results in under 12 hours, even across thousands of audio samples and multiple languages. With this new funding, we’ll expand our engineering team, strengthen post-analysis tooling, and launch in key global markets including the U.S., Europe, and Southeast Asia. A huge thank you to our investors Serac Ventures, Naver D2SF, and KAIST Ventures for believing in our vision to become the trust layer for voice AI. We are just getting started. 🚀

[Why Post-Refining Matters in Voice AI] Running large-scale evaluations is no longer the bottleneck. The real challenge? Making sense of all the data you’ve collected. Raw scores are just the beginning. Without post-refinement, you risk missing critical insights and making decisions on shaky ground. ✅ Which raters were inconsistent? ✅ Were results stable across languages or age groups? ✅ Are you measuring true quality, or just loudness and accent? At Podonos, we go beyond the numbers. We help you turn raw evaluation results into clear, structured insights. So you can build, compare, and deploy voice models with confidence. 👉 Stop guessing. Start refining. 📖 Read the full story: https://lnkd.in/g_Pzwbzv

Voice AI evaluation often focuses on metrics like naturalness, similarity, or emotion. But one critical factor is often overlooked: Who is evaluating your model, and how were they selected? Poor evaluator quality leads to: ▪️ Misinterpreted criteria ▪️ Low language proficiency ▪️ Noisy environments or bad devices ▪️ Inconsistent or careless scoring That’s why prescreening matters. At Podonos, every evaluator goes through a rigorous vetting process: ▪️ Language fluency tests ▪️ Audio equipment and environment checks ▪️ Scoring consistency on test samples ▪️ Attention to detail through controlled tasks Only those who pass are allowed to evaluate. ✅ What you get: ▪️ Cleaner, more trustworthy feedback ▪️ Actionable insights ▪️ Confidence in what the scores really mean When your Voice AI is judged by the right people, you make the right decisions. 🔗 Learn how Podonos do it: https://lnkd.in/gd_igca3 #VoiceAI #HumanEvaluation #AIQuality #EvaluationMatters #Podonos #SpeechTechnology #Prescreening #AIethics

Celebrating the 10th anniversary of Google for Startups Korea, ✦Soohyun Bae sat down for an inspiring conversation with the vibrant and friendly Hannah Curran on campus. Huge thanks to the GFS team—Podonos is growing ultra-fast with your incredible support!

Expanding your Voice AI model to multiple languages is no longer optional—it's essential. Perfecting your model in just one language isn’t enough to reach and resonate with global audiences. But adding new languages can be one of the most demanding challenges. Podonos offers a seamlessly designed evaluation solution that saves time and effort through precise measurement and rapid turnaround. Discover how Podonos can help you scale globally—and why multilingual evaluation makes all the difference in the link below. 👇 Podonos blog: https://lnkd.in/gMwpJQzz

Google released Gemini 2.5 TTS model, and Podonos measured its real performance by comparing it to ElevenLabs' model. And we are sharing the analysis report in public! 📝 Overview of Gemini 2.5 Flash VS ElevenLabs Multilingual V2 ☑️ Overall preferences show that ElevenLabs and Gemini are similar, and ElevenLabs slightly wins over Gemini. ☑️ Both are weak at pronouncing addresses and numbers. ☑️ Gemini is weaker at dialogs pronouncing celebrity names, medical terms. ☑️ Gemini's diversity & inclusion is weak: Better at male voices and weak at female voices. Find more insights in the full report in the comment below. 👇

"No to guesstimate! Yes to measure!" That’s exactly what Resemble AI did when launching Chatterbox, their latest text-to-speech model. Instead of relying on vague impressions, they used Podonos to benchmark performance with real data. Show the true quality of your Voice AI—backed by transparent, reliable evaluation. Podonos accelerates model improvement and delivers deep insights to guide your critical business decisions. 🥸 Check our Case Study in the comment below. 👇

🎉Big congratulations to the Resemble AI team for unlocking the next wave of Voice AI innovation by opensourcing Chatterbox. This is a truly monumental moment, getting worldwide attention with a groundbreaking product. We’re happy to see Chatterbox released to the world, backed by Podonos' trusted and transparent evaluation.

🚀 We’re thrilled to see Resemble AI officially opensources their new TTS model, Chatterbox! Designed with standout strengths in naturalness, Chatterbox is now live on GitHub and HuggingFace. Before the release, Resemble AI used Podonos to benchmark Chatterbox against ElevenLabs — ensuring the results were accurate, trustworthy, and delivered fast. We’re proud to have powered the evaluation behind this transparent and confident launch. 🔗 Check the comments for a link to Resemble AI’s benchmark results.👇

Podonos' product gets improved each day as we speak! 🚀 We recently launched our newly designed website, and we’re excited to share it with you. Whether you're a partner of Podonos or just discovering us, we invite you to take a look around - link in the comment. 👇 We’d love to hear what you think. Feedback is always welcome.