Project Goal

  • Use the latest AI/ML tools and technologies available to automate Audiobook Quality Control (QC) tasks to save time and money.
  • Reduce time spent on QC tasks from 2-3 days to a few minutes.
  • A system that can be used by PRH divisions in all territories, not just the US.


Project Description

Orion QC (Quality Control) is a system to automate the testing and validation process of Audio Books produced by the PRH Audio Production Group. This system will supplement the existing manual Proofreading and QC process.

It is difficult for a human to pay attention to every detail, so letting a computer take over the tasks is beneficial.



Manual QC

Orion QC



Unique, custom-built system. (There is nothing like this out there!)

Manually listen to the audiobook and compares it to the manuscript

Does not lead to fatigue, no human errors

Proofread / QC in a few minutes



Manuscript: PDF or eBook

(partial support)

Manuscript: Text format

Audio Files: WAV / MP3 / FLAC

Proofreading / QC tasks


Detect missing content

Identify additional or repeated content

Detect misread audio content

Detect long silence (over 3 sec) 

 ✅ (difficult)

Detect Incidental noise and digital glitches

❌  (roadmap)



Structured Excel document with discrepancies

Problem location as a timestamp in the audio file

Surrounding text at the problem location (before and after)

Color-coded HTML file to clearly display discrepancies

Play a short clip of the discrepancy


Project Dates

Research and Proof of concept 2021

Alpha release Q3 2021

Beta release Q1 2022



PRH US IT Content Applications Group: 

          * Srinivasan Muruganandam

          * Justin Branco

          * Dzmitry Kasinets

PRH US Audio Production Group: 

          * Ok Hee Kolwitz


Top 3 Benefits

  • Takes a few minutes to QC hours of Audio Books, compared to 2-3 days for manual QC for a 5-hour audiobook
  • Cost savings for the company
  • Can proofread in any of the 37 languages supported by Amazon Transcribe


Top 3 Challenges

  • Identifying true differences (that the audio team really cares about) between the manuscript and the audio file. Minimize false negatives/differences.
  • Reference document / Manuscript text does not "exactly" match the output of Amazon Transcribe service. For example, the manuscript may have a date of 2/24/2022 and the audio transcript may have February 24, 2020, and the application should not show them as a difference
  • Text matches should be exact, not fuzzy

Further Resources

More information about this project (Please contact me if you are unable to access the file)

Amazon Transcribe