Skip to content

Python tool for converting files and office documents to Markdown.

Add support for conversion via Document Intelligence (#303)

KennyZhang1bfde8574204d
diff --git a/pyproject.toml b/pyproject.toml
index 9c113ade..2a4e203e 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -42,6 +42,8 @@ dependencies = [
"pathvalidate",
"charset-normalizer",
"openai",
+ "azure-ai-documentintelligence",
+ "azure-identity"
]
[project.urls]
diff --git a/src/markitdown/__main__.py b/src/markitdown/__main__.py
index b6cf963b..69e8f0ea 100644
--- a/src/markitdown/__main__.py
+++ b/src/markitdown/__main__.py
@@ -4,8 +4,8 @@
import argparse
import sys
from textwrap import dedent
-from .__about__ import __version__
-from ._markitdown import MarkItDown, DocumentConverterResult
+from __about__ import __version__
+from _markitdown import MarkItDown, DocumentConverterResult
def main():
@@ -57,16 +57,37 @@ def main():
"--output",
help="Output file name. If not provided, output is written to stdout.",
)
+ parser.add_argument(
+ "-d",
+ "--use-docintel",
+ action="store_true",
+ help="Use Document Intelligence to extract text instead of offline conversion. Requires a valid Document Intelligence Endpoint.",
+ )
+ parser.add_argument(
+ "-e",
+ "--endpoint",
+ type=str,
+ help="Document Intelligence Endpoint. Required if using Document Intelligence.",
+ )
args = parser.parse_args()
- if args.filename is None:
+ if args.use_docintel:
+ if args.endpoint is None:
+ raise ValueError(
+ "Document Intelligence Endpoint is required when using Document Intelligence."
+ )
+ elif args.filename is None:
+ raise ValueError("Filename is required when using Document Intelligence.")
+ markitdown = MarkItDown(docintel_endpoint=args.endpoint)
+ else:
markitdown = MarkItDown()
+
+ if args.filename is None:
result = markitdown.convert_stream(sys.stdin.buffer)
- _handle_output(args, result)
else:
- markitdown = MarkItDown()
result = markitdown.convert(args.filename)
- _handle_output(args, result)
+
+ _handle_output(args, result)
def _handle_output(args, result: DocumentConverterResult):
diff --git a/src/markitdown/_markitdown.py b/src/markitdown/_markitdown.py
index 33806e13..ae6a7b4f 100644
--- a/src/markitdown/_markitdown.py
+++ b/src/markitdown/_markitdown.py
@@ -33,6 +33,19 @@ import requests
from bs4 import BeautifulSoup
from charset_normalizer import from_path
+# Azure imports
+from azure.ai.documentintelligence import DocumentIntelligenceClient
+from azure.ai.documentintelligence.models import (
+ AnalyzeDocumentRequest,
+ AnalyzeResult,
+ DocumentAnalysisFeature,
+)
+from azure.identity import DefaultAzureCredential
+
+# TODO: currently, there is a bug in the document intelligence SDK with importing the "ContentFormat" enum.
+# This constant is a temporary fix until the bug is resolved.
+CONTENT_FORMAT = "markdown"
+
# Optional Transcription support
IS_AUDIO_TRANSCRIPTION_CAPABLE = False
try:
@@ -1318,6 +1331,74 @@ class ZipConverter(DocumentConverter):
)
+class DocumentIntelligenceConverter(DocumentConverter):
+ """Specialized DocumentConverter that uses Document Intelligence to extract text from documents."""
+
+ def __init__(
+ self,
+ endpoint: str,
+ api_version: str = "2024-07-31-preview",
+ ):
+ self.endpoint = endpoint
+ self.api_version = api_version
+ self.doc_intel_client = DocumentIntelligenceClient(
+ endpoint=self.endpoint,
+ api_version=self.api_version,
+ credential=DefaultAzureCredential(),
+ )
+
+ def convert(
+ self, local_path: str, **kwargs: Any
+ ) -> Union[None, DocumentConverterResult]:
+ # Bail if extension is not supported by Document Intelligence
+ extension = kwargs.get("file_extension", "")
+ docintel_extensions = [
+ ".pdf",
+ ".docx",
+ ".xlsx",
+ ".pptx",
+ ".html",
+ ".jpeg",
+ ".jpg",
+ ".png",
+ ".bmp",
+ ".tiff",
+ ".heif",
+ ]
+ if extension.lower() not in docintel_extensions:
+ return None
+
+ # Get the bytestring for the local path
+ with open(local_path, "rb") as f:
+ file_bytes = f.read()
+
+ # Certain document analysis features are not availiable for filetypes (.xlsx, .pptx, .html)
+ if extension.lower() in [".xlsx", ".pptx", ".html"]:
+ analysis_features = []
+ else:
+ analysis_features = [
+ DocumentAnalysisFeature.FORMULAS, # enable formula extraction
+ DocumentAnalysisFeature.OCR_HIGH_RESOLUTION, # enable high resolution OCR
+ DocumentAnalysisFeature.STYLE_FONT, # enable font style extraction
+ ]
+
+ # Extract the text using Azure Document Intelligence
+ poller = self.doc_intel_client.begin_analyze_document(
+ model_id="prebuilt-layout",
+ body=AnalyzeDocumentRequest(bytes_source=file_bytes),
+ features=analysis_features,
+ output_content_format=CONTENT_FORMAT, # TODO: replace with "ContentFormat.MARKDOWN" when the bug is fixed
+ )
+ result: AnalyzeResult = poller.result()
+
+ # remove comments from the markdown content generated by Doc Intelligence and append to markdown string
+ markdown_text = re.sub(r"<!--.*?-->", "", result.content, flags=re.DOTALL)
+ return DocumentConverterResult(
+ title=None,
+ text_content=markdown_text,
+ )
+
+
class FileConversionException(BaseException):
pass
@@ -1337,6 +1418,7 @@ class MarkItDown:
llm_model: Optional[str] = None,
style_map: Optional[str] = None,
exiftool_path: Optional[str] = None,
+ docintel_endpoint: Optional[str] = None,
# Deprecated
mlm_client: Optional[Any] = None,
mlm_model: Optional[str] = None,
@@ -1406,6 +1488,12 @@ class MarkItDown:
self.register_page_converter(ZipConverter())
self.register_page_converter(OutlookMsgConverter())
+ # Register Document Intelligence converter at the top of the stack if endpoint is provided
+ if docintel_endpoint is not None:
+ self.register_page_converter(
+ DocumentIntelligenceConverter(endpoint=docintel_endpoint)
+ )
+
def convert(
self, source: Union[str, requests.Response, Path], **kwargs: Any
) -> DocumentConverterResult: # TODO: deal with kwargs