Simultaneous experience of the same acoustic stimulus in two distinct phenomenological modes, e.g. as a speech-like and as a non-speech event, is referred to as duplex perception (DP). The most widely investigated DP paradigm splits each of the stop consonant-vowel (CV) syllables /ga/ and /da/ into an isolated formant transient (chirp) and the remaining sound structure (base). The present study recorded mismatch fields in response to a series of dichotically applied base and chirp components using whole-head magnetencephalography (MEG). Preattentive mismatch fields showed larger amplitudes in response to contralateral deviants. During attention to the fused percept /da/, the left ear deviants chirps elicited an enhanced and posteriorly shifted dipole field over the ipsilateral hemisphere. These data provide first neurophysiological evidence that the integration of acoustic stimulus elements into a coherent syllable representation constitutes a distinct stage of left-hemisphere speech sound encoding.