Merge pull request #566 from pipecat-ai/mb/make-markdown-modifiable

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

CHANGELOG.md

@@ -9,9 +9,9 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ### Added
 
-- Added `MarkdownRemovalProcessor`. This processor removes markdown formatting
-  from a TextFrame. It's intended to be used between the LLM and TTS in order
-  to remove markdown from the text the TTS speaks.
+- Added a new util called `MarkdownTextFilter` which is a subclass of a new
+  base class called `BaseTextFilter`. This is a configurable utility which
+  is intended to filter text received by TTS services.
 
 - Added new `RTVIUserLLMTextProcessor`. This processor will send an RTVI
   `user-llm-text` message with the user content's that was sent to the LLM.

src/pipecat/services/ai_services.py

@@ -37,6 +37,7 @@
 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 @@ def __init__(
         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 @@ def __init__(
         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 @@ async def _update_settings(self, settings: Dict[str, Any]):
                 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 @@ async def _push_tts_frames(self, text: str):
             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:

src/pipecat/utils/text/base_text_filter.py

@@ -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

src/pipecat/utils/text/markdown_text_filter.py

@@ -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