Whisper Normalization

A Python package for text normalization, specifically designed for use with Whisper models.

Note: This code is extracted from OpenAI's Whisper repository to allow for standalone usage of the text normalization modules without the heavy dependencies of the full Whisper package.

• Original Source: openai/whisper
• License: MIT

Installation

You can install this package directly from source:

pip install .

You can also be installed from PyPi:

pip install whisper-normalization

Usage

from whisper_normalization import EnglishTextNormalizer, BasicTextNormalizer

# English Normalization
normalizer = EnglishTextNormalizer()
text = "Mr. Smith bought $5.50 worth of apples in the 1990s."
normalized = normalizer(text)
print(normalized)
# Output: "mister smith bought five dollars and fifty cents worth of apples in the nineteen nineties"

# Basic Normalization
basic_normalizer = BasicTextNormalizer()
text = "Bonjour à tous!"
normalized = basic_normalizer(text)
print(normalized)

Features

• Basic text cleaning (symbol removal, diacritic handling)
• Advanced English normalization:
  • Number to text conversion (integers, decimals, currencies, years)
  • Contraction expansion
  • Abbreviation expansion
  • British to American spelling normalization