This repository accompanies “Evaluating Strategies for Multimodal Bidirectional Interactions between Pedestrians and AVs.” It contains the materials needed to inspect and reproduce the paper’s online-questionnaire statistics and controlled speech-pipeline benchmarks.
questionnaire/: LimeSurvey instrument, original survey media, anonymized quantitative and open-ended responses, coding audit, analysis code, and generated tables.benchmarks/: sanitized latency measurements, evaluation audio, ASR references and hypotheses, analysis code, and generated summaries.docs/: data dictionary.
Python 3.9 or newer is required.
python3 -m venv .venv
source .venv/bin/activate
python -m pip install -r requirements.txt
python run_all.pyThe command regenerates questionnaire/results/ and benchmarks/results/, then checks the headline values and released-data integrity. It does not download data or contact external services.
The quantitative survey table contains only new participant IDs, broad demographics, ordinal responses, ranks, and categorical pairing selections. LimeSurvey response IDs, dates, timing telemetry, recruitment-source fields, and profession were removed. Open-ended responses are released separately with the same randomized IDs; internal author annotations were removed, while participant text was otherwise preserved.
The authors' original repository material is licensed under GPL-3.0. Third-party material retains its own terms, including CC BY 4.0 for FLEURS and CC BY-NC 3.0 for UrbanSound8K-derived audio. See LICENSE, LICENSES/README.md, and THIRD_PARTY.md. Citation metadata are in CITATION.cff.