Dual-Modality Dataset of English-Chinese Simultaneous Interpreting at the 80th UN Anniversary High-Level Plenary: Speech Features and Text Quality

It aims to address the unimodal bias in existing diplomatic SI assessment by providing integrated dual-modality data of "speech features + text quality".The dataset includes 46 minutes and 30 seconds of valid audio (44.1kHz/16bit, .wav format) and aligned text, segmented into 90 analysis fragments from three professional UN interpreters.

Speech data was preprocessed via Audacity (SNR ≥40dB) and annotated using Praat, covering 6 core speech features (e.g., disfluency rate, intonation alignment, breath control) with inter-annotator consistency ≥95%. Text data was transcribed (Chinese accuracy ≥95%) and evaluated by expert raters across 5 quality dimensions (information integrity, linguistic accuracy, cultural-affective equivalence, pragmatic appropriateness, textual fluency), with inter-annotator reliability Cohen’s κ=0.89.Supplementary files include source language feature annotations (terminology density, syntactic complexity), interpreter strategy classifications (efficiency-, accuracy-, stability-oriented), and statistical analysis protocols.