ARPAbets

ARPAbets are a set of ASCII phoneme symbols used to transcribe American English for speech processing tasks. The term derives from ARPA (the Advanced Research Projects Agency) and refers to the ARPAbet phoneme set that became widely adopted in early speech recognition research and in the CMU Pronouncing Dictionary. ARPAbet provides a simple, machine-friendly way to encode speech sounds without relying on the International Phonetic Alphabet.

The notation divides phonemes into consonants and vowels, using two-letter or longer codes. Vowels in ARPAbet are typically followed by a digit (0, 1, or 2) to indicate stress on the syllable in which the phoneme occurs; this stress information is particularly useful for speech recognition and synthesis systems. ARPAbet is designed for American English and is commonly used in pronunciation dictionaries and acoustic models. It is most notably associated with the CMU Pronouncing Dictionary, which maps English words to sequences of ARPAbet symbols.

In practice, ARPAbet serves as a bridge between written text and phonetic representations in many speech technologies. It is often used in traditional hidden Markov model-based pipelines, as well as in some hybrid and neural systems that still rely on explicit phoneme sequences. The notation is not identical to the IPA and is primarily a tool for computational processing rather than a universal phonetic standard.

Example: the word HELLO is typically transcribed as HH AH0 L OW1 in ARPAbet, illustrating the use of consonant and vowel symbols with a primary stress on the final syllable.