Speech produced in noisy environments (Lombard speech) is characterized by a range of acoustic and phonetic changes. These changes stem from increased speaking effort which reflects communicative intent as well as decreased auditory feedback of the speaker’s own voice. An accurate understanding of real-world Lombard effects is important in hearing science for the development and assessment of signal processing strategies targeting realistic speech signals. While Lombard effects are well known from the literature, studies of Lombard speech have typically been based on relatively unnatural speaking tasks such as reading from a script and have been measured in simplified acoustic backgrounds such as stationary noise or constructed babble noise. Lombard speech produced under such unnatural conditions may differ significantly from speech produced in real-world settings. This study describes a novel method of eliciting natural conversational speech across five highly realistic everyday acoustic environments. Through the increased realism of both the speaking task and acoustic backgrounds this study aims to provide a more ecologically valid approximation of real-world Lombard speech than has been previously reported. Based on recordings of conversations between 10 pairs of young, normal-hearing people, a continuum of ordered acoustic and phonetic changes in speech is described in relation to changes in acoustic environments and is related to self-reported listening effort ratings across acoustic environments.