Blog
Page 2 of 22
Measuring benchmark optimization in speech recognition
New research introduces three tests to quantify benchmark optimization in speech recognition, finding top open-source ASR models reproduce benchmark transcripts even when the audio contradicts them.
Theo Lebryk, Eric Bezzam, Alice Baird, David Ayllon, Jakub Piotr Cłapa, Jens Madsen, Panagiotis Tzirakis / Aug 21, 2026
Introducing Real World VoiceEQ: Measuring the Human Quality of Voice AI
Existing benchmarks suggest voice AI is nearing human-level performance but real-world conversations tell a different story.
Alice Baird / Jul 14, 2026
Emotional Intelligence Is a Training-Time Property, Not a Prompt
You can't prompt your way to emotional intelligence — perception and expression live in the weights, not the system prompt. The scarce asset is the human-grounded reward signal that trains them in.
Andrew Ettinger / Jul 1, 2026
Speech-to-Speech Is the Hardest Problem in Voice AI
Speech-to-speech models promise to break the ceiling of the transcript-based voice pipeline. The architecture is right — but promise isn't production, and demos aren't evidence. Measurement is the deciding input.
Andrew Ettinger / Jun 1, 2026
Disentangling Emotion from Voice: A Cross-Product Sampling Approach for Expressive Voice Data
Humans can separate how they feel from how they speak. Voice models still struggle to do the same.
Hoon Shin / May 27, 2026
The Science of What a Voice Reveals
Emotion isn't six categories — it's a continuous, high-dimensional space, carried as much by the voice as the words. Measuring it correctly is the foundation voice AI is built on.
Andrew Ettinger / May 4, 2026