Infants are known to have more difficulties detecting speech in the presence of a noisy background compared to adults. Most of the previous studies used behavioral methods to assess this capacity, whereas electroencephalography (EEG) may provide objective neural markers of this ability. Our study aims to assess infants' speech detection abilities in a noisy background by examining the auditory event related potentials (ERP) using a passive listening procedure.

We recruited a group of 25 three-month-old infants who were following-up at 10 months of age. We also recruited a group of 31 adults for comparisons. All listeners were awake throughout the EEG recordings and were presented with the syllables "ba" or "da" in a continuous stationary speech-shaped noise. The syllables were delivered via two speakers in front of the listeners and were randomly played at four different signal-to-noise ratios (SNR = +3, -4, -8, -15 dB). A total of 52 trials per SNR was played while measuring ERPs for about 20 min. The ERP amplitude after syllable onset was averaged over five frontal electrodes and in the time window 400:600 ms, corresponding to the P300-like ERP response.

A first analysis showed that for both infant groups and the adult group, only ERP amplitude obtained at the SNR -15 dB were not significantly different from the neural responses obtained during the noise inter-trial presentation. A linear-mixed model was then run on the mean [400-600ms] ERP amplitude and revealed a significant effect of Age (p = .002), SNR level (p < .001), syllables (p <.001) and interaction between Age and SNR (p = .006). Post-hoc analyses indicated that at 3 months of age, decreasing the SNR significantly decreased the ERP amplitude, while at 10 months of age the ERP amplitude differed significantly between the -15 and +3, -15 and -4, and -8 and +3 conditions, but not between any other pairs.

These findings showed in both infant groups an effect of SNR on the P300-like amplitude: the greater SNRs, the greater the ERP amplitude. This result is in line with adult studies showing that the P300 amplitude decreases as the difficulty of discriminating stimuli increases. Importantly, ERP response were significantly different from neural noise from SNR equal to -8 dB suggesting that as for the adults, the infants’ brain is able to detect syllable at this SNR level. Our results are promising for objective clinical measures of speech in noise.