Misperception in Northern Vietnamese Tones due to the Lombard effect

Studies examining Lombard speech have found that intelligibility often increases due to acoustic changes such as elevated intensity, increased F0 and F0 range, and elongated sonorous segments. By previously studying the production of the eight lexical tones in Northern Vietnamese in noise, we have also observed that all tones experienced differing degrees of the Lombard effect. For instance, tone B2 (mid-falling with creakiness) showed resistance to increased glottal regularity, while tone C2 (mid-rising with creakiness) exhibited a greater convergence towards regular phonation. The current study conducted a perception experiment to investigate to what extent those acoustic adaptations due to hyper-articulation impact listeners' perception of tonal identity. Native Vietnamese listeners identified the tones produced in quiet or in noise of 78 or 90 dB SPL by a male and a female speaker. The results show that the creaky tones B2 and C2 had the lowest identification accuracy while the rising modal tone B1 received the highest. Tone C2 in particular was most likely to be identified as the high tones A1 or B1 when originally uttered in noise. The findings call into question whether hyper-articulation always leads to better speech intelligibility, especially for tonal languages.