Someone says "I'm fine." You know they aren't. You heard it - the pause, the flat tone. Voice AI doesn't. It all gets dropped at transcription. So we built the Expression Measurement API. 48+ emotions, 600+ descriptors, 50+ languages, audio and video, real-time. Conversation was always more than words.

Text-to-speech was built to read a single voice aloud. Now it's being asked to hold a conversation between two, and that's a different problem entirely. The speakers have to stay distinct, trade turns with natural timing, and keep their emotional states from bleeding into each other. None of that shows up in word error rate. So we built a rubric: five dimensions, 48 two-speaker scripts, and four scenarios designed to break a dialogue by pairing similar-sounding voices, forcing frequent interruptions, contrasting emotions, and running long exchanges. Three blind human raters per clip. We ran three generations of Google's Gemini TTS models through it, alongside other multi-speaker systems. Keeping two voices apart is the hardest part of the job. Every system handles a male and female pair easily and drops on at least one same-gender pairing. The newest Flash model made its biggest gain on exactly that. Models with the same overall score get there differently. One leads on keeping speakers’ voices distinct, another on making the dialogue sound like one continuous recording. The overall score hides that difference. New capabilities need new evaluations. Full case study, with audio examples, is linked in the comments.

New scores are up on the Real World VoiceEQ text-to-speech board, plus two new scoring metrics and a new voice controllability leaderboard.

Congratulations to the @OpenAI team on the GPT-Live-1 launch. Here is the early read from our RWVoiceEQ STS evals. Voice naturalness is where it moves: up about 70% on their previous flagship, from near the bottom to level with Gemini 3.1 Flash Live at the top. What was OpenAI's weakest factor is now their strongest. Emotion understanding also improved, about 10% up and the highest score in the set. Emotion alignment is where it doesn't. Whether the model notices when tone contradicts the words, a hesitant or frustrated yes to a critical confirmation: unchanged from the previous flagship, under 1% apart. Sounding human and understanding humans are separate problems. The industry is solving the first one quickly. The second one is where the work is. Measuring the gap is how we close it. Early numbers. Full leaderboard update to follow.

We asked 11 voice cloning models to copy the same 25 voices, then asked real people: is that the same person? The most natural-sounding clone ranked 8th of 11 on sounding like the person. New from Hume AI: the Voice Replication Leaderboard. Live now on @huggingface, links below.

Today, voice models have no problem generating “angry” or “sad” expressions. But ask for: → bored + fast → joy + shy → disappointment + confident …and most systems collapse into stereotypes. Our latest research blog explores why this happens — and how disentangling emotion from voice at the data layer improves expressive control. Read more below!

Use OpenAI's new open source model with any voice, cloned or designed!

2024: Voice Cloning 2025: What about personality cloning? Hume’s voice AI can now not only mimic your voice but also speaking style and language. It’s now available via our TTS and new speech-to-speech model, EVI 3, which is also launching today.

Octave’s first-of-its-kind voice intelligence speaks for itself 😉 In a blind study, Octave outperformed ElevenLabs Voice Design: 🔊71.6% preferred Octave’s audio quality 🗣️51.7% found Octave more natural 🎯57.7% said Octave better matched voice descriptions