This study investigated the effects of expanding the gross time-amplitude variations of ‘speech-envelope noise’ stimuli on speech recognition. The initial stimuli were VCV logatomes presented in quiet or against a steady white noise with a 0-dB signal-to-noise ratio. Their low-frequency temporal modulations (<500 Hz) were extracted in broad frequency bands, and raised to the power 2. The resulting envelopes were then used to modulate a white noise, and combined to produce the ‘speech-envelope noise’ stimuli. As a consequence, listeners were forced to identify speech using primarily temporal envelope cues. The results obtained with four normal-hearing listeners show small decrements in recognition performance of 1–15% when expanding the envelope of the speech stimuli presented in quiet. The results also show a small but consistent improvement in performance of 6–14% when expanding the envelope of the speech stimuli presented in noise. These results are consistent with those obtained by Fu and Shannon (J. Acoust. Soc. Am. 104 (1998) 2570–2577) with speech presented in quiet. They also suggest that the reduction in the modulation depth of the speech envelope caused by noise or reverberation could be compensated by expanding low-frequency temporal modulations.