Existing literature has demonstrated that individuals with autism spectrum disorder (ASD) exhibit atypical use of contextual information in their surroundings. However, there is limited understanding regarding their integration of contextual cues in speech processing. This study aims to explore how Mandarin-speaking children with and without ASD identify lexical tones in speech and nonspeech contexts, and to determine whether the size of context effect would be modulated by children’s cognitive abilities. Twenty-five children with ASD and 25 typically developing (TD) children were asked to identify Mandarin lexical tones preceded by three types of contexts (speech, nonspeech, and nonspeech-flattened contexts). We also tested child participants’ verbal intelligence, nonverbal intelligence, and working memory capacity. Results revealed that the context effect was only observed in the speech contexts, where Mandarin-speaking children with ASD exhibited a reduced context effect compared to TD children. Moreover, TD children with higher verbal intelligence demonstrated a diminished context effect. However, nonverbal intelligence and working memory capacity were not significantly associated with the size of context effect in either group. These findings revealed a subtle yet important difference between ASD and TD children’s utilization of speech contexts in lexical tone identification, and validated a speech-specific mechanism underpinning children’s lexical tone normalization.