Understanding the neural mechanisms underlying speech processing by the auditory system holds paramount importance in our comprehension of how we communicate, of the challenges of learning a new language, and of potential therapies for disorders of speech communication. Despite the wealth of knowledge gained from studies in human subjects (Yi et al., 2021; Oganian et al., 2023), the intricate neural processes involved in speech processing remain challenging to explore fully. Ethical considerations and technical limitations restrict the depth of insight that can be obtained solely through research in humans. A full understanding of how the brain processes and categorizes the acoustic features of speech requires descriptions of how neurons represent these features and how neural connections change when learning novel acoustic categories.

Because this level of investigation is extremely challenging to achieve in human subjects, an appropriate animal model is needed (Kluender, 2000; Lotto et al., 2003). Previous studies have demonstrated that various animal species are capable of learning phonetic categories that share perceptual qualities with humans (Kuhl and Padden, 1983; Kuhl and Miller, 1978; Engineer et al., 2015; Saunders and Wehr, 2019), suggesting that animal models are appropriate for the study of human speech processing by the auditory system. In particular, the mouse provides an unparalleled level of experimental access to investigate how the activity of individual neurons represents distinct acoustic features and how the circuits these neurons form change during the learning process. However, acoustic communication in mice is markedly different from that in humans. Mouse vocalizations are ultrasonic, the repertoire of “syllables” is more limited, and there is evidence that mouse vocalization is not learned, as vocalizations emitted by deaf mice do not differ in either structure or usage from hearing mice (Portfors, 2007; Portfors and Perkel, 2014; Hammerschmidt et al., 2012; Mahrt et al., 2013). Here, we assessed whether the mouse can serve as a good model organism for studying the neural mechanisms of speech processing, and evaluated whether neurons in distinct regions of the mouse auditory cortex are sensitive to specific features of speech.

We report that mice are able to learn to categorize frequency-shifted human speech sounds that varied according to two features: (1) formant transitions (FT), the spectral change in formant frequencies during a consonant (e.g., the difference between /ba/ and /da/); and (2) voice onset time (VOT), the period of time from the burst of a plosive to the onset of the vocal fold vibration (e.g., the difference between /ba/ and /pa/). Moreover, we show that neurons across many auditory cortical areas of the mouse (including primary, dorsal, and ventral auditory cortex, as well as temporal association area) are selective to these acoustic features, with the dorso-posterior region of the auditory cortex containing a higher proportion of speech-selective neurons compared to other areas. These combined behavioral and physiological results demonstrate that the mouse can serve as a valid model organism for investigating the encoding of speech features by the auditory system and how these neural representations may change throughout learning.