Data and Supplemental Materials for Noise-Robust Voice Command Recognition for Human-Robot Collaboration in Construction Environments

<p dir="ltr">This dataset contains the research data and supplemental materials supporting the manuscript "Noise-Robust Voice Command Recognition for Human-Robot Collaboration in Construction Environments." It includes (1) 1,000 synthetic Mandarin construction-oriented human-robot collaboration commands recorded as 16-kHz mono WAV files, together with speaker-profile-disjoint train, validation, and test splits, command metadata, voice-design prompts, selected synthetic reference voices, and quality-control summaries; (2) 2,586 10-s construction-noise clips derived from field recordings, totaling 7.18 h, with fixed split assignments and acoustic descriptors; (3) the structured semantic-command parser prompt; (4) the supplemental tables and figures; and (5) a video demonstrating representative robot-motion simulation and safety-control responses.

The command speech was generated with Qwen3-TTS. Speaker role, age-range, and gender attributes are synthesis controls rather than observations of real construction workers. The construction-noise clips were screened to remove incidental speech and unrelated events.