Previous studies have suggested that the perception of vowels and consonants changes from language-universal to language-specific between 6 and 12 months of age. This report suggests that language-specific perception emerges even earlier for lexical tones. Experiment 1 tested English-learners' perception of Cantonese tones, replicating declines in tone discrimination from 4 to 9 months of age. Experiment 2 tested infants learning "non-native" versus "native" tone systems (Mandarin-learners versus Cantonese-learners). All Chinese-learners discriminated the tones, but showed language-specific differences in tone "preferences" at both ages. Indeed, English-, Mandarin-, and Cantonese-learning 4-month-olds all exhibited distinct preferences. With other work, this shows that language-specific speech perception emerges over a more complex and extended schedule than previously thought: first for lexical stress and tone (less than 5 months), then vowels (6-8 months), consonants (8.5-12 months), and finally phoneme duration (18 months). Acoustic salience likely plays an important role in determining the timing of phonetic development. (Contains 6 figures and 2 tables.)