wx44wx/pipecat
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Mark the Markdown processor a util, and allow it to take inputs

Browse Source
This commit is contained in:
Mark Backman
2024-10-10 15:16:47 -04:00
parent 5b8753c8b6
commit cbecae40a9
5 changed files with 109 additions and 92 deletions
Download Patch File Download Diff File Expand all files Collapse all files

92
src/pipecat/processors/text/markdown_remover.py
View File

@@ -1,92 +0,0 @@
#
# Copyright (c) 2024, Daily
#
# SPDX-License-Identifier: BSD 2-Clause License
#
import re
from markdown import Markdown
from pydantic import BaseModel
from pipecat.frames.frames import Frame, TextFrame
from pipecat.processors.frame_processor import FrameDirection, FrameProcessor
class MarkdownRemovalProcessor(FrameProcessor):
"""Removes Markdown formatting from text in TextFrames.
Converts Markdown to plain text while preserving the overall structure,
including leading and trailing spaces. Handles special cases like
asterisks and table formatting.
"""
class InputParams(BaseModel):
speak_code: bool = True
def __init__(self, params: InputParams = InputParams(), **kwargs):
super().__init__(**kwargs)
self._md = Markdown()
self._params = params
async def process_frame(self, frame: Frame, direction: FrameDirection):
await super().process_frame(frame, direction)
if isinstance(frame, TextFrame):
cleaned_text = self._remove_markdown(frame.text)
await self.push_frame(TextFrame(text=cleaned_text))
else:
await self.push_frame(frame, direction)
def _remove_markdown(self, markdown_string: str) -> str:
# Replace newlines with spaces only when there's no text before or after
markdown_string = re.sub(r"^\s*\n", "", markdown_string, flags=re.MULTILINE)
# Remove backticks from inline code, but not from code blocks
markdown_string = re.sub(r"(?<!`)`([^`\n]+)`(?!`)", r"\1", markdown_string)
# Remove repeated sequences of 5 or more characters
markdown_string = re.sub(r"(\S)(\1{4,})", "", markdown_string)
# Preserve numbered list items with a unique marker, §NUM§
markdown_string = re.sub(r"^(\d+\.)\s", r"§NUM§\1 ", markdown_string)
# Preserve leading/trailing spaces with a unique marker, §
# Critical for word-by-word streaming in bot-tts-text
preserved_markdown = re.sub(
r"^( +)|\s+$", lambda m: "§" * len(m.group(0)), markdown_string, flags=re.MULTILINE
)
# Convert markdown to HTML
extensions = ["fenced_code"] if self._params.speak_code else []
md = Markdown(extensions=extensions)
html = md.convert(preserved_markdown)
# Remove code blocks if speak_code is False
if not self._params.speak_code:
html = re.sub(r"<code>.*?</code>", "", html, flags=re.DOTALL)
# Remove HTML tags
text = re.sub("<[^<]+?>", "", html)
# Replace HTML entities
text = text.replace("&nbsp;", " ")
text = text.replace("&lt;", "<")
text = text.replace("&gt;", ">")
text = text.replace("&amp;", "&")
# Remove leading/trailing asterisks
# Necessary for bot-tts-text, as they appear as literal asterisks
text = re.sub(r"^\*{1,2}|\*{1,2}$", "", text)
# Remove Markdown table formatting
text = re.sub(r"\|", "", text)
text = re.sub(r"^\s*[-:]+\s*$", "", text, flags=re.MULTILINE)
# Restore numbered list items
text = text.replace("§NUM§", "")
# Restore leading and trailing spaces
text = re.sub("§", " ", text)
return text

7
src/pipecat/services/ai_services.py
View File

@@ -37,6 +37,7 @@ from pipecat.processors.frame_processor import FrameDirection, FrameProcessor
from pipecat.transcriptions.language import Language
from pipecat.utils.audio import calculate_audio_volume
from pipecat.utils.string import match_endofsentence
from pipecat.utils.text.base_text_filter import BaseTextFilter
from pipecat.utils.time import seconds_to_nanoseconds
from pipecat.utils.utils import exp_smoothing
@@ -172,6 +173,7 @@ class TTSService(AIService):
stop_frame_timeout_s: float = 1.0,
# TTS output sample rate
sample_rate: int = 16000,
text_filter: Optional[BaseTextFilter] = None,
**kwargs,
):
super().__init__(**kwargs)
@@ -182,6 +184,7 @@ class TTSService(AIService):
self._sample_rate: int = sample_rate
self._voice_id: str = ""
self._settings: Dict[str, Any] = {}
self._text_filter: Optional[BaseTextFilter] = text_filter
self._stop_frame_task: Optional[asyncio.Task] = None
self._stop_frame_queue: asyncio.Queue = asyncio.Queue()
@@ -242,6 +245,8 @@ class TTSService(AIService):
self.set_model_name(value)
elif key == "voice":
self.set_voice(value)
elif key == "text_filter" and self._text_filter:
self._text_filter.update_settings(value)
else:
logger.warning(f"Unknown setting for TTS service: {key}")
@@ -312,6 +317,8 @@ class TTSService(AIService):
return
await self.start_processing_metrics()
if self._text_filter:
text = self._text_filter.filter(text)
await self.process_generator(self.run_tts(text))
await self.stop_processing_metrics()
if self._push_text_frames:

0
src/pipecat/processors/text/__init__.py → src/pipecat/utils/text/__init__.py
View File

18
src/pipecat/utils/text/base_text_filter.py Normal file
View File

@@ -0,0 +1,18 @@
#
# Copyright (c) 2024, Daily
#
# SPDX-License-Identifier: BSD 2-Clause License
#
from abc import ABC, abstractmethod
from typing import Any, Mapping
class BaseTextFilter(ABC):
@abstractmethod
def update_settings(self, settings: Mapping[str, Any]):
pass
@abstractmethod
def filter(self, text: str) -> str:
pass

84
src/pipecat/utils/text/markdown_text_filter.py Normal file
View File

@@ -0,0 +1,84 @@
#
# Copyright (c) 2024, Daily
#
# SPDX-License-Identifier: BSD 2-Clause License
#
import re
from typing import Any, Mapping
from markdown import Markdown
from pydantic import BaseModel
from pipecat.utils.text.base_text_filter import BaseTextFilter
class MarkdownTextFilter(BaseTextFilter):
"""Removes Markdown formatting from text in TextFrames.
Converts Markdown to plain text while preserving the overall structure,
including leading and trailing spaces. Handles special cases like
asterisks and table formatting.
"""
class InputParams(BaseModel):
enable_text_filter: bool = True
def __init__(self, params: InputParams = InputParams(), **kwargs):
super().__init__(**kwargs)
self._settings = params
def update_settings(self, settings: Mapping[str, Any]):
for key, value in settings.items():
if hasattr(self._settings, key):
setattr(self._settings, key, value)
def filter(self, text: str) -> str:
if self._settings.enable_text_filter:
# Replace newlines with spaces only when there's no text before or after
text = re.sub(r"^\s*\n", " ", text, flags=re.MULTILINE)
# Remove repeated sequences of 5 or more characters
text = re.sub(r"(\S)(\1{4,})", "", text)
# Preserve numbered list items with a unique marker, §NUM§
text = re.sub(r"^(\d+\.)\s", r"§NUM§\1 ", text)
# Preserve leading/trailing spaces with a unique marker, §
# Critical for word-by-word streaming in bot-tts-text
preserved_markdown = re.sub(
r"^( +)|\s+$", lambda m: "§" * len(m.group(0)), text, flags=re.MULTILINE
)
# Convert markdown to HTML
md = Markdown()
html = md.convert(preserved_markdown)
# Remove HTML tags
filtered_text = re.sub("<[^<]+?>", "", html)
# Replace HTML entities
filtered_text = filtered_text.replace("&nbsp;", " ")
filtered_text = filtered_text.replace("&lt;", "<")
filtered_text = filtered_text.replace("&gt;", ">")
filtered_text = filtered_text.replace("&amp;", "&")
# Remove double asterisks (consecutive without any exceptions)
filtered_text = re.sub(r"\*\*", "", filtered_text)
# Remove single asterisks at the start or end of words
filtered_text = re.sub(r"(^|\s)\*|\*($|\s)", r"\1\2", filtered_text)
# Remove Markdown table formatting
filtered_text = re.sub(r"\|", "", filtered_text)
filtered_text = re.sub(r"^\s*[-:]+\s*$", "", filtered_text, flags=re.MULTILINE)
# Restore numbered list items
filtered_text = filtered_text.replace("§NUM§", "")
# Restore leading and trailing spaces
filtered_text = re.sub("§", " ", filtered_text)
return filtered_text
else:
return text