voice-agent-examples / src /nvidia_pipecat /services /blingfire_text_aggregator.py
fciannella's picture
Working with service run on 7860
53ea588
raw
history blame
2.26 kB
#
# Copyright (c) 2024–2025, Daily
#
# SPDX-License-Identifier: BSD 2-Clause License
#
"""Blingfire-based text aggregator for sentence detection."""
import blingfire as bf
from loguru import logger
from pipecat.utils.text.base_text_aggregator import BaseTextAggregator
class BlingfireTextAggregator(BaseTextAggregator):
"""This is a text aggregator that uses blingfire for sentence detection.
It aggregates text until a complete sentence is detected using blingfire's
sentence segmentation capabilities.
"""
def __init__(self):
"""Initialize the BlingfireTextAggregator with an empty text buffer."""
self._text = ""
logger.debug("BlingfireTextAggregator: Initialized new instance")
@property
def text(self) -> str:
"""Return the currently aggregated text.
Returns:
str: The text currently being aggregated.
"""
return self._text
async def aggregate(self, text: str) -> str | None:
"""Aggregate text and return a complete sentence when detected.
Args:
text (str): The text to be aggregated.
Returns:
str | None: A complete sentence if one is detected, otherwise None.
"""
result: str | None = None
self._text += text
logger.debug(f"BlingfireTextAggregator: Aggregating text: '{self._text}'")
# Use blingfire to split text into sentences
sentences_text = bf.text_to_sentences(self._text)
sentences = [s.strip() for s in sentences_text.split("\n") if s.strip()]
# If we have multiple sentences, return the first complete one
if len(sentences) > 1:
result = sentences[0]
if self._text.lstrip().startswith(sentences[0].strip()):
self._text = self._text[len(sentences[0]) :]
return result
async def handle_interruption(self):
"""Handle interruption by clearing the aggregated text buffer."""
logger.debug("BlingfireTextAggregator: Handling interruption")
self._text = ""
async def reset(self):
"""Reset the aggregated text buffer to empty."""
logger.debug("BlingfireTextAggregator: Resetting text buffer")
self._text = ""