buchwandler

English API

English G2P provides high-quality phoneme conversion for US and British English.

Semantic Preparation

English written-to-spoken preparation is owned by the shared source-aligned spokenform path, with lexical abbreviation recognition and customization owned by the shared abbr2words registry. Once Spokenform accepts a source span, kokorog2p preserves its semantic replacement and provenance. Dates, times, currencies, quantities, temperatures, safe ordinary numbers, and exact decimal digit reading are prepared once per homogeneous language run, with protected source spans preserved.

Kokorog2p retains English typography normalization, tokenization, lexicon, G2P and fallback behavior. The English NumberConverter remains downstream for reserved and phoneme-sensitive forms such as years, ordinals, Roman numerals, phone/version-like sequences, and suffix transformations. Token normalization is typography-only after source-aligned semantic preparation.

Main Class

Lexicon

Number Conversion

Converter Class

Helper Functions

Constants

Examples

Basic Usage

from kokorog2p.en import EnglishG2P

# US English
g2p = EnglishG2P(language="en-us")
tokens = g2p("Hello world!")

for token in tokens:
    print(f"{token.text} -> {token.phonemes}")

# British English
g2p_gb = EnglishG2P(language="en-gb")
tokens = g2p_gb("Hello world!")

spaCy Model Selection

English G2P uses spaCy for POS tagging when use_spacy=True. You can choose the spaCy English model with spacy_model:

from kokorog2p.en import EnglishG2P

# Default model (recommended balance)
g2p_md = EnglishG2P(use_spacy=True, spacy_model="en_core_web_md")

# Smaller model
g2p_sm = EnglishG2P(use_spacy=True, spacy_model="en_core_web_sm")

# Larger model
g2p_lg = EnglishG2P(use_spacy=True, spacy_model="en_core_web_lg")

Dictionary Lookup

from kokorog2p.en import EnglishLexicon

lexicon = EnglishLexicon(language="en-us")

# Simple lookup
phonemes = lexicon.lookup("hello")
print(phonemes)  # həlˈO

# POS-aware lookup
read_present = lexicon.lookup("read", tag="VB")
read_past = lexicon.lookup("read", tag="VBD")

Number Expansion

from kokorog2p.en import EnglishG2P

# Numbers are automatically expanded during G2P processing
g2p = EnglishG2P(language="en-us")
tokens = g2p("I have $42.50 and 3 cats.")

for token in tokens:
    print(f"{token.text} -> {token.phonemes}")
# → I -> aɪ
# → have -> hæv
# → forty-two dollars and fifty cents -> ...
# → and -> ænd
# → three -> θɹi
# → cats -> kæts

Punctuation Normalization

English G2P automatically normalizes punctuation variants:

from kokorog2p.en import EnglishG2P

g2p = EnglishG2P(language="en-us")

# Apostrophe variants (all normalize to ')
g2p("don't")    # Right single quote (')
g2p("don't")    # Apostrophe (')
g2p("don`t")    # Grave accent (`)
g2p("don´t")    # Acute accent (´)

# Ellipsis variants (all normalize to …)
g2p("Wait...")       # Three dots
g2p("Wait. . .")     # Spaced dots
g2p("Wait…")         # Ellipsis character

# Dash variants (all normalize to — when spaced)
g2p("Wait - now")    # Hyphen with spaces
g2p("Wait -- now")   # Double hyphen
g2p("Wait – now")    # En dash
g2p("Wait — now")    # Em dash
g2p("Wait ― now")    # Horizontal bar
g2p("Wait ‒ now")    # Figure dash
g2p("Wait − now")    # Minus sign

# Compound words keep hyphens (then removed in output)
g2p("well-known")         # Hyphen joins words
g2p("state-of-the-art")   # Multiple hyphens

Normalized Characters:

  • Apostrophes: ' ' ' `` `` ´ ʹ ′ ＇ → '

  • Ellipsis: ... . . . .. .... → …

  • Dashes (when spaced): - -- – ― ‒ − → —