Content The dataset contains processed audio files employed in a listening test performed at the Here East Audio Lab of the University College London to derive a model of acoustic perception in residential buildings [1]. The study followed a full factorial design by combining five urban environments (Factor A) and four indoor sound scenarios (Factor B). The audio files are available in both B-format (Ambix) and A-format. Factor A (Outdoor sounds) Factor A audio recordings were performed in indoor spaces without audible indoor sound sources and with windows partially opened to different urban contexts in the city of London. Sound material was recorded @24bit/48kHz using a First Order Ambisonics (FOA) microphone (Sennheiser AMBEO VR Mic) positioned at the average listener’s ear height in endfire position, with accompanying portable multi-channel audio recorder (Sound Devices MixPre-10T) with channels 1-4 linked for the FOA setting, together with a sound level meter (NTi Audio XL2), both microphones oriented towards the window openings. By recording outdoor acoustic environments in indoor spaces, the effects of reverberation and window filtering were intrinsically embedded in the collected recordings. Factor B (Indoor sounds) Factor B audio recordings were made in low-noise indoor environments using the equipment described above with both microphones oriented roughly towards the sound source of interest. Combinations of Factors A & B Excluding the two “no added sounds” scenarios, a total of seven audio stimuli were played and combined during the listening tests, as described in [1], resulting in total 20 scenarios where no more than 2 sounds were overlapped. Audio Editing and Processing Audio samples were edited and processed in A format in the Digital Audio Workstation Reaper (Cockos) @24bit/48kHz. The edits were performed in terms of removing extraneous sound events by trimming the audio track and creating the necessary crossfades, in order to bring the audio material as close as possible to the scenario it represented. Audio processing was conducted using the Sennheiser Ambeo plugin to generate the B-format audio files, to be correctly spatialized using a playback system of choice. In the process of conversion to B format, the default Ambisonics Correction Filter was engaged and the Low Cut Filter was switched off, while the Microphone Rotation was set to correct for the endfire position used during the recordings. One-minute excerpts were finally extracted. No further audio editing, nor processing was done. Full details about sound recordings and playback levels used in the experiment are available in [1] and in the supplementary materials. The audio files are intended to be employed in future listening tests. [1] Torresin, S., Albatici, R., Aletta, F., Babich, F., Oberman, T., Siboni, S., & Kang, J. (2020). Indoor soundscape assessment: A principal components model of acoustic perception in residential buildings. Building and Environment, 182, 107152.