University significantly improved transcription speed through implementation of an AI speech recognition model

Summary:

The AI Model designed for the University is capable of transforming audio or video into 100% accurate text and subtitling, reducing transcription time from three weeks to just 5-7 working days.

Client: 

Amsterdam University of Applied Sciences (HVA)

Problem Statement: 

The CAREM research team at Amsterdam University of Applied Sciences (HVA) conducts various types of research, often involving recorded interviews. These recordings need to be transcribed for use in their studies. However, making audio and video content widely accessible by converting it into text was previously unfeasible due to the substantial effort needed to manually create transcripts or subtitles.

Results: 

• Decreased transcription time from 3 weeks to 5-7 working days.

• Transforming audio or video into 100% accurate text and subtitling.

• AI speech recognition model is specifically trained for a particular language.

• The hourly rate of transcription remained more-less the same. 

• The new workflow enables researchers to concentrate on the tasks that are most important to them.

AI Solution Overview:

The Amsterdam University of Applied Sciences has adopted Amberscript’s transcription services to convert recorded interviews into text. Amberscript develops speech recognition models specifically trained for particular languages. For example, it handles the subtitling of more than half of Dutch municipal council meetings using an engine tailored to the political language, resulting in highly accurate transcriptions.

Supported by the Ministry of Economic Affairs and Climate Policy, Amberscript is developing a speech recognition model specifically for universities. These models are also customized for individual customers. AI and speech recognition technology expedite this process and reduce costs. Amberscript helps Amsterdam University of Applied Sciences save the time and effort required for manual transcription, delivering 100% accurate text and subtitles. Their models, designed for European languages, achieve the highest market accuracy, closely matching human transcription levels.

Amberscript also offers an online text processor. Users can upload audio or video files, after which the audio is converted into text. Users or Amberscript’s transcription staff can then correct any errors, achieving 100% accuracy. The corrections made in the text processor are stored as training data, which gradually reduces errors in automatic speech recognition.

The entire transcription process is streamlined with Amberscript’s AI engine performing the initial transcription at 85%-95% accuracy, resulting in a clean-read transcription. Each source file undergoes a manual check by a transcriber to reach near-perfect accuracy (99%-100%). Finally, source files are uploaded to an online editor for a final review.

References: 

1. Amberscript: making audio accessible for the deaf and hard of hearing. https://nlaic.com/en/use_cases/amberscript-making-audio-accessible-for-the-deaf-and-hard-of-hearing/
2.  Amberscript: On a mission to make all audio accessible. https://aimagazine.com/technology/amberscript-mission-make-all-audio-accessible
3. Amberscript – Automated speech translation for public organisations. https://ai-watch.github.io/AI-watch-T6-X/service/90077.html
4. How Amsterdam University of Applied Sciences improved transcription speed significantly. https://www.amberscript.com/en/academy/hva-case-study/

Industry: Educational Services

Vendor: Amberscript

Client: Amsterdam University of Applied Sciences

Publication Date: 2022