Implementation:Speechbrain Speechbrain G2P Tool

Knowledge Sources	SpeechBrain
Domains	Text_Processing, G2P
Last Updated	2026-02-09 00:00 GMT

Overview

Concrete tool for grapheme-to-phoneme conversion using a pretrained G2P model provided by the SpeechBrain library.

Description

This is a command-line convenience script that transcribes text into phoneme sequences using a pretrained Grapheme-to-Phoneme (G2P) model loaded via speechbrain.inference.text.GraphemeToPhoneme. It supports three operating modes: (1) interactive shell mode for exploratory analysis where users can enter text line by line, (2) single-text mode for transcribing a single example from the command line, and (3) file mode for batch-processing text files of arbitrary size with results written to an output file. The file mode processes inputs in configurable batch sizes with progress tracking. Phoneme output uses a space-separated format with <spc> tokens for word boundaries. The G2P models can be trained using the recipes located in recipes/LibriSpeech/G2P.

Usage

Use as a CLI tool for converting orthographic text to phoneme sequences. Particularly useful for preprocessing text datasets for TTS (Text-to-Speech) training where phoneme inputs are preferred over raw graphemes.

Code Reference

Source Location

• Repository: SpeechBrain
• File: tools/g2p.py

Signature

def transcribe_text(g2p, text):
    """Transcribes a single line of text and outputs it."""
    ...

def transcribe_file(g2p, text_file_name, output_file_name=None, batch_size=64):
    """Transcribes a file with one example per line."""
    ...

def get_line_count(text_file_name):
    """Counts the lines in a file without loading it into memory."""
    ...

def transcribe_stream(g2p, text_file, output_file, batch_size=64, total=None):
    """Transcribes a file stream in batches."""
    ...

def chunked(iterable, batch_size):
    """Break iterable into lists of length batch_size."""
    ...

Import

# Interactive mode
python g2p.py --model /path/to/model --interactive

# Single text
python g2p.py --model /path/to/model --text "Hello world"

# File mode
python g2p.py --model /path/to/model --text-file input.txt --output-file phonemes.txt

I/O Contract

Inputs

Name	Type	Required	Description
--model	str	Yes	Path to the pretrained G2P model directory
--hparams	str	No	Name of the hyperparameter file (default: hyperparams.yaml)
--text	str	No	Single text string to transcribe
--text-file	str	No	Path to a text file with one sample per line
--output-file	str	No	Path for output file (stdout if omitted)
-i, --interactive	flag	No	Launch an interactive transcription shell

Outputs

Name	Type	Description
phonemes	str	Space-separated phoneme sequence with <spc> as word boundary
output_file	file	File containing phoneme transcriptions, one per line

Usage Examples

# Start an interactive G2P shell
python g2p.py --model /path/to/g2p_model --interactive
# > Enter text: The cat sat on the mat
# DH AH <spc> K AE T <spc> S AE T <spc> AA N <spc> DH AH <spc> M AE T

# Transcribe a single sentence
python g2p.py --model /path/to/g2p_model --text "This is a line of text"

# Batch transcribe a file for TTS preprocessing
python g2p.py --model /path/to/g2p_model \
    --text-file sentences.txt \
    --output-file phonemes.txt

Related Pages

• Principle:Speechbrain_Speechbrain_ASR_Evaluation_With_WER