While Microsoft Office is often considered to be the “truth”, there are times when different versions of Office disagree on how the document should look, a problem which is particularly likely with older documents.

The variation in the way that different tools render the same DOCX file is a major reason for converting them to PDF – increasing the consistency of appearance across a wide range of tools and platforms.

Manually converting documents to PDF using Microsoft Word is not, however, a scalable approach. It works for five documents but not for 500,000.

That is where an automated Office-to-PDF conversion solution becomes necessary.

Automating Microsoft Word is possible, but Word was not designed for unattended server-side automation, which can cause problems, for example, if a modal dialog appears then the conversion process will hang.

LibreOffice, on the other hand, was designed with automated conversion in mind. This is done by calling it in “headless” mode, and offers a cheap, open-source option.

Apryse Server SDK also offers a way to convert DOCX files to PDF, but it is both closed source and has a license fee.

In this article we will look at why using Apryse Server SDK is a strong case despite the free alternatives.

When evaluating a tool, you should of course use files that are representative of the types of documents that you use in your business.

Since that varies from industry to industry, for this article we will use files taken from two publicly accessible corpora.

The first corpus was DOCXCORP which has over 736,000 DOCX files. There is a range of languages, and the corpus is catalogued by subject areas, including Legal, Education, Government, and Healthcare. 100 documents were “blindly” chosen from the corpus (rather than cherry-picking documents that would favor the Apryse Server SDK). The selection took every one hundredth file starting at #42.

The second set of documents are from the GovDocs1 corpus which contains almost a million documents collected from government web servers in a wide range of formats, but which are around fifteen years old. All of the DOC and DOCX files from one folder (#042) were used (which led to some interesting results which are described later)

The testing environment was a Windows 11 PC with 13th Gen i7-1365U processor, 16GB RAM, 10 Cores and 12 Logical Processors.

In production, you could use a dedicated Windows or Linux server. Production hardware affects processing speed and which fonts are available.

The Apryse Server SDK supports the use of a substitute font package which can significantly improve font fidelity. Read more about font substitution when converting a DOCX file to PDF.

For this article, no dedicated font substitution file was used for the Apryse Server SDK, so conversion relied on locally available fonts.

The following tools were used:

Microsoft Word was controlled programmatically.

Headless LibreOffice was controlled via a script using:

Apryse Server SDK was controlled via a script which ultimately called the following:

Note: LibreOffice was found to work unreliably in a multi-threaded environment, so after several attempts the processing was performed by a single threaded app.

Apryse Server SDK worked reliably when called from both a single and a multi-threaded harness.

Microsoft Word for Microsoft 365 was used to produce the baseline PDF. Fidelity therefore means similarity to the output from that specific Word version in the stated test environment.

Document fidelity was calculated using the Apryse Server SDK to compare the converted output against the baseline created by Word. This was performed in a C# app.

Using the Apryse Server SDK may appear "circular" at first, however comparing two different PDFs is not remotely the same as converting a Word document into a PDF, and being able to do both with a single SDK is a strength not a weakness.

Measurements included page count accuracy, font accuracy, the number of text differences that were detected, and so on. These measurements are simple to calculate, fast to collect, and relevant to output fidelity.

The importance of each specific aspect depends on your use-case, for example, the correct page count often matters more than whether the number of words on a specific page remains the same.

Results were calculated for each pair of PDFs and then aggregated across each dataset.

If you would like to use the comparison tool with your own documents, then please contact us at blog-feedback@apryse.com.

The Apryse Semantic Text difference mechanism, is great, but it is not perfect. Sometimes finding differences that are not visually significant.

Post-process custom code was added to remove pairs of edits where a deletion was immediately followed by an insertion of the same text, and where codepoints for characters changed but the result was non-visible to the user (for example the way in which bullets can be encoded).

Semantic text differences across page boundaries were retained.

The ability to work with data extracted from PDFs at this level is one of the great things about working with the Apryse Server SDK.

As noted earlier, attempts to use LibreOffice in a multi-threaded (or multi-process) environment failed. As such, LibreOffice conversions were only performed single-threaded.

Conclusion Apryse Server SDK is 6x faster than LibreOffice for older doc and docx files, and up to 14x faster for modern files. Even in a single threaded environment the conversion is 2.5 to 6 times faster than for LibreOffice

If you are only converting a single file, then the difference between using the Apryse Server SDK and LibreOffice will save you perhaps 3 seconds. But if you had to convert 10,000 documents, then that relates to a difference of 8 hours.

The testing mechanism measures a number of features. Only the main ones are included in this article.

Each document gets an overall score from 0 to 100 based on four sub-scores:

The overall score is calculated by weighting these sub-scores: text 0.30, layout 0.25, font 0.25, and page 0.20.

The number of identified differences was slightly higher for Apryse than for LibreOffice with this set of documents, but looking deeper, many of those differences were found in a single file, where a fixed-width font and a very specific right-hand margin caused lines to break at different words. Each change was considered a difference, even though the document remained perfectly readable - it is very much an edge case that had been included in the test-set by chance.

It would have been easy to exclude that file from the test-set, but the whole intention was to allow the superior quality of Apryse conversions to be shown with a "random" set of files, not a cherry-picked one as used by some companies.

Both Apryse and LibreOffice sometimes used fonts that did not match those used by Microsoft Word to create the “truth” PDF. This affected 13% of the files converted by Apryse but 21% of those converted by LibreOffice.

Even though no attempt was made to get Apryse to use a dedicated font mapping process, it offers far more consistent results than LibreOffice.

Font substitution can cause either trivial aesthetic changes in the appearance of the document, or, if the font metrics differ significantly, this can affect the number of words that fit on a page, resulting in changes in page layout, and with text overflowing from one page to another causing extra pages to be created.

Since font substitution can affect page layout, the test was repeated with just the subset of 79 documents for which LibreOffice had not substituted fonts.

Effectively, that involved cherry picking files that LibreOffice does well with. Even with that set, Apryse gave consistently higher fidelity.

The older sample set illustrated two problems that are common with real-world data.

Apryse and LibreOffice, on the other hand, were able to convert these four files. The data from these is not included in the results however since there was no baseline available to compare the conversion with.

Once again, Apryse gave consistently higher fidelity conversions.

LibreOffice provides a free, open-source way to automate Office-to-PDF conversion. However, in our Windows test environment, Apryse provided substantially higher throughput and more consistently reproduced the page count, fonts, and content of the PDFs generated by Microsoft Word.

The size of the performance difference depended on both the document set and the execution model. Apryse was approximately 2.4 to 7.0 times faster in the reported single-threaded tests. Because Apryse also ran reliably with concurrent conversions in our harness, its measured corpus-level throughput advantage increased to approximately 5.8 times for the older files and 14.8 times for the modern files.

Conversion fidelity also favored Apryse across the reported page-count, font, and overall scoring measures. The results do not mean that every document will convert better with Apryse, and performance will vary with hardware, fonts, document complexity, and software versions. They do indicate that for this corpus and test configuration, Apryse offered the stronger combination of throughput, operational reliability, and similarity to the Microsoft Word baseline.

To wrap up, Apryse offers faster, higher fidelity conversion of DOCX files to PDF than is available using LibreOffice.

If you have any questions, you can reach out to us on Support channel.

If you have any questions about this blog, or there are other blogs that you would like us to write, then please contact us at blog-feedback@apryse.com.