Bangla Text Normalizer

A robust, dependency-free Bangla Text Normalizer (Bengali Text Normalizer) Python package designed for NLP (Natural Language Processing) and TTS (Text-to-Speech) pipelines. It accurately converts numeric symbols, dates, time, currencies, phone numbers, and abbreviations into their spoken Bangla textual format.

Unlike other tools, this package handles the Indian Numbering System (Lakh/Crore) correctly and provides context-aware normalizations for improved audio data quality.

This tool is optimized for:

• Text-to-Speech (TTS) Preprocessing: Converting raw text into perfectly pronounceable Bangla words.
• ASR & NLP Data Preparation: Standardizing text for training machine learning models.
• Chatbots & Voice Assistants: Ensuring textual responses are ready for voice synthesis.

Installation

pip install bangla-text-normalizer

Usage

from bangla_text_normalizer import BanglaTextNormalizer

normalizer = BanglaTextNormalizer()

# Example 1: Addresses and Numbers
text = "৫ম তলা, ৩/বি, গুলশান-২, ঢাকা ১২১২"
print(normalizer.normalize(text))
# Expected Output:
# পঞ্চম তলা, ৩/বি, গুলশান-২, ঢাকা এক দুই এক দুই

# Example 2: Dates (Smart Pronunciation)
print(normalizer.normalize("21/02/1952")) 
# Output: একুশে ফেব্রুয়ারি ঊনিশশো বাহান্ন

# Example 3: Phone Numbers
print(normalizer.normalize("01711000000"))
# Output: শূন্য এক সাত এক এক শূন্য শূন্য শূন্য শূন্য শূন্য শূন্য

# Example 4: Decimals and Fractions
print(normalizer.normalize("3.5"))
# Output: তিন দশমিক পাঁচ

Features

• Number Conversion: Converts integers to text (Supports extremely large numbers like Lakhs/Crores).
• Date Normalization:
  • Handles DD/MM/YYYY formats.
  • Context-aware dates: 21 -> একুশে (Ekmshe), 01 -> পহেলা (Pohela).
  • Smart Year handling: 1952 -> ঊনিশশো বাহান্ন (Nineteen Hundred Fifty-Two), 2024 -> দুই হাজার চব্বিশ.
• Phone Numbers:
  • Normalizes BD phone numbers (e.g., 017..., +88017...) into individual digits.
  • Robust handling of spaces and hyphens (e.g., 017 11000000).
• Ordinals:
  • Supports numeric ordinals: 1st, 2nd, 10th.
  • Supports Bangla short-forms: ১ম, ২য়, ১০ম.
  • Auto-generates ordinals for large numbers: 1023rd -> এক হাজার তেইশতম.
• Decimals & Fractions: Handles floating point numbers (3.1416) and simple fractions (1/2).
• Zero Dependency: Pure Python implementation with no external requirements.

License

MIT License