AI Paper Atlas

Research map / Large language models / Language, safety and interpretability

TTSDS2: Resources and Benchmark for Evaluating Human-Quality Text to Speech Systems

Christoph Minixhofer, Ondrej Klejch, Peter Bell

ICLR 2026 · Large language models › Language, safety and interpretability · approach: Speech and audio language models · 12 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
use TTSDS2 instead of original TTSDSTTSDS metric · English average over 4 domainsSpearman correlation with subjective scores: 0.67 average, surpassing the original by 10% relativeImproved
use multilingual feature extractors (mHuBERT-147, XLSR-53) in TTSDS2TTSDS2 · Uriel+ typological distanceSpearman correlation: -0.39 -> -0.51Improved
replace WER features with ASR final-layer activations for IntelligibilityTTSDS · real-speech scores across domainsscore robustnessImproved
replace HuBERT token-length prosody features with speaking rateTTSDS · real-speech scoresscore on real speechImproved
remove ESC-50 from noise distractor datasets (ablation)TTSDS2 · preliminary experimentsscore consistencyImproved
use Speaker factor alone vs other factors (ablation)TTSDS2 factors · CleanPearson correlation with MOS: Speaker 0.84 vs Generic 0.42, Prosody 0.38, Intelligibility 0.47Mixed
use Speaker factor alone vs other factors (ablation)TTSDS2 factors · NoisyPearson correlation with MOS: Speaker 0.86 vs Generic 0.59, Prosody 0.46, Intelligibility 0.59Mixed
use Generic factor alone vs Speaker (ablation)TTSDS2 factors · KidsPearson correlation with MOS: Generic 0.80 vs Speaker 0.70Mixed
use Speaker factor alone (ablation)TTSDS2 factors · WildPearson correlation with MOS: Speaker 0.59 vs Generic 0.53, Intelligibility 0.58, Prosody 0.34Mixed
compare TTSDS2 with X-Vector speaker similarity on Kids MOSTTSDS2 · Kids MOSSpearman correlation: 0.61 vs X-Vector 0.70No clear effect
compare TTSDS2 with UTMOS on Wild MOSTTSDS2 · Wild MOSSpearman correlation: 0.75 vs UTMOS -0.12Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy