🔡ARPABET

ARPABET (also spelled ARPAbet) is a set of ASCII codes for the phonemes of General American English. It was developed by ARPA in the 1970s as part of the Speech Understanding Research (SUR) project, and it does one specific job well: it lets you write the sounds of English words using only plain keyboard characters — AA, AE, IY, UW, CH, JH, with digits for stress. The single most important thing to understand about it is what it is not: it is a phoneme code for one language, not a general phonetic alphabet like the IPA.¹²³

Two systems — and the one everyone uses

ARPABET was defined in two forms: a one-character system (each sound is a single character, using alternating upper and lower case) and a one-or-two-character, case-insensitive system. The two-letter, case-insensitive form is far more widely adopted — it is the version used in the CMU Pronouncing Dictionary — and in it, segments are separated by spaces.²⁴

Stress digits attach to vowels only

In the CMUdict tradition, a digit right after a vowel marks its stress: 0 = unstressed, 1 = primary, 2 = secondary. So the CMUdict entry for "natural" is N AE1 CH ER0 AH0 L. The digits go on vowels only — never on consonants — which is a common slip when generating ARPABET.²

Base ARPABET vs the TIMIT variant

A revised ARPABET is used in the TIMIT speech corpus, and it is not the same set. TIMIT adds symbols the base set lacks — most notably stop-closure symbols (BCL, DCL, GCL, KCL, PCL, TCL, marking the silent closure before a plosive release), voiced-h HV (ɦ), syllabic consonants (ENG and friends), a devoiced schwa AX-H, epenthetic silence EPI, a pause PAU, and a begin/end marker H#. The authoritative list is the LDC's TIMIT PHONCODE.TXT. Guides and code must not conflate base ARPABET with the TIMIT variant.²⁵

Where it lives

ARPABET's living home is the CMU Pronouncing Dictionary, the standard free pronunciation lexicon for US English, which keeps the system in daily use across automatic speech recognition, text-to-speech, and NLP pipelines. It was also used historically in several speech synthesizers (SAM, the Amiga "Say" utility, Votrax-based devices).³²

Errors to avoid

• Using ARPABET for non-English data — it is English-only by design.²⁶
• Treating ARPABET codes as IPA-equivalent narrow transcription — it is a phoneme code (plus a few allophones), not narrow phonetics.⁶
• Conflating base ARPABET with the TIMIT variant (closures, HV, syllabics).⁴⁵
• Putting stress digits on consonants — they attach to vowels only.²
• Confusing ARPABET with TIMITBET or WorldBet — the WorldBet memo treats all three as distinct sets.⁷

Learn more

• The CMU Pronouncing Dictionary — The living reference: US-English pronunciations in two-letter ARPABET with stress digits.
• ARPABET (Wikipedia) — full phone tables — One- and two-letter tables plus the TIMIT-variant table.
• WorldBet — our companion guide — The multilingual ASCII scheme that treats Arpabet and TIMITBET as distinct sets.
• The IPA — our companion guide — The general phonetic alphabet ARPABET is often wrongly equated with.