NeMo Text Processing

1.1.0 verified Mon Apr 27 auth: no python

NVIDIA NeMo text processing toolkit for ASR and TTS, providing text normalization (TN) and inverse text normalization (ITN) for over 20 languages. Current version 1.1.0 (August 2024). Release cadence is irregular with minor version bumps every few months.

pip install nemo-text-processing

Common errors

error ImportError: cannot import name 'get_normalizer' from 'nemo_text_processing.text_normalization' ↓

cause The function was moved to nemo_text_processing.utils in v1.0.0.

fix

Change import to: from nemo_text_processing.utils import get_normalizer

error pynini.Arc - no such symbol error during normalization ↓

cause Incompatible pynini version (e.g., pynini 2.1.6 with Python 3.12).

fix

Install pynini via conda-forge or downgrade Python to 3.10/3.11.

Warnings

gotcha The normalizer expects input_case='cased' or 'lower_cased'; using default may produce unexpected results for certain texts. ↓

fix Explicitly set input_case parameter.

breaking In v1.0.0, the get_normalizer function moved from nemo_text_processing.text_normalization to nemo_text_processing.utils. ↓

fix Update import to from nemo_text_processing.utils import get_normalizer

gotcha Normalization depends on pynini (WFST library). Installation issues on Windows or Python 3.12+ may cause runtime errors. For Windows, consider installing pynini from conda-forge. ↓

fix Use conda install -c conda-forge pynini on Windows, or use a Linux environment.

Imports

NEMO_TN_EN

wrong

from nemo_text_processing.tn import NEMO_TN_EN

correct

from nemo_text_processing.text_normalization.en.utils import NEMO_TN_EN

Incorrect deep import path (tn vs en)

Normalizer
```
from nemo_text_processing.text_normalization.normalize import Normalizer
```
Direct import for English normalizer; else use NEMO_TN_EN or NEMO_ITN_EN

get_normalizer

wrong

from nemo_text_processing.text_normalization import get_normalizer

correct

from nemo_text_processing.utils import get_normalizer

Changed from text_normalization to utils in recent releases

Quickstart

Basic text normalization using English normalizer

from nemo_text_processing.text_normalization.normalize import Normalizer
normalizer = Normalizer(input_case='cased', lang='en')
text = "He was born on Dec 3rd, 1978."
normalized = normalizer.normalize(text, verbose=True)
print(normalized)
# Output: "He was born on December third, nineteen seventy eight."