We released the inference code and model of a part of AESCA (Yamamoto+, #ASRU2025), the top-performing system in AudioMOS Challenge Track 2, to predict the audio aesthetics score (AES).
Paper: arxiv.org/abs/2512.05592
Code: github.com/CyberAgentAI...
At the IEEE #ASRU2025, we presented our automatic evaluation system for generated audio, which won first place in the AudioMOS Challenge 2025 Track 2🥇. At the start of the session, an award ceremony was held, and I accepted the certificate on behalf of the team.
Today’s poster presentation at #ASRU2025 🥳
Preprint: arxiv.org/abs/2512.05592
On Dec 9th, 4:00 PM, we will be giving a poster presentation titled “The T12 System for AudioMOS Challenge 2025: Audio Aesthetics Score Prediction System Using KAN- and VERSA-based Models” at ASRU2025 in Honolulu. #ASRU2025
Preprint: arxiv.org/abs/2512.05592
We are attending #ASRU2025 in Honolulu!!!🏝️🌺 The conference center is very close to Waikiki beach 🌊🏄🌈
LLM Metric Scores Self‑Supervised Speech Models Without Training
An LLM‑based metric scores speech models via log‑likelihood of token sequences, avoiding extra training. Accepted to the 2025 IEEE ASRU conference, it showed high correlation with ASR benchmark. getnews.me/llm-metric-scores-self-s... #asru2025 #llmeval
