Case study — Media Monitoring

The first media monitoring company in Romania, approved as a FIBEP member since 2004

mediaTRUST is the largest media monitoring organization in Romania. Its primary function is to monitor the image of individuals, brands, businesses, and institutions.

mediaTRUST's monitoring services include print, web, radio, and television, encompassing a team of over 60 experts, who know everything there is to know about media. The organization has already conducted hundreds of studies on various brands and companies.

mediaTRUST

The Challenge

Finding a more accurate and more affordable solution.

Achieving a higher level of accuracy compared to Google's speech-to-text solution, and finding a new speech recognition solution with a high degree of scalability, able to handle large volumes of audio data. Finding a more affordable solution, as the other solutions they had tried were expensive and with questionable results.

mediaTRUST had three main objectives:

  • Create highly accurate transcripts with the highest possible accuracy rate and monitor brand and product mentions as effectively as possible.
  • Be able to always deliver the most accurate transcription to the final customers.
  • Transcribe all the company names, product names, brand names, terminology, and specific words or mentions, in order for their final customers to analyze when the conversation is about a specific product, brand, person or company.

The Solution

Tailored speech models that perform better than big tech.

mediaTRUST tested the infrastructure's accuracy and scalability by performing extensive load testing and simultaneously sending thousands of files for transcription. All of the tests were highly successful. With Vatis Tech, they were able to train and customize a speech model using audio from mediaTRUST's own platform, which led to a dramatic accuracy increase beyond what the all-in-one ASRs could reach. mediaTRUST achieved:

  • A speech-to-text API for the Romanian language with over 95% accuracy rate on media type of audio data, and a transcription speed of approximately 15% of the file length — 1 hour of audio was transcribed in less than 10 minutes.
  • Improvements on specific keywords, primarily brands or products.
  • An integration process that was ready in less than a day, with the remaining time spent only on performing accuracy and scalability tests.
  • A successful implementation process, due to our API integration, which was something challenging, if not impossible, with other providers. Due to our deployment flexibility, mediaTRUST was able to install our solution in their own infrastructure.
mediaTRUST

Vatis provides us with a high-quality, fast, efficient, and high-standard product. They are extremely quick in communication, which is essential for us. We can rely on them to deliver highly accurate transcriptions for large volumes of linguistic data. Moreover, it is a pleasure to work with them.

Bogdan ApostolCEO, mediaTRUST

The Results

  • Cost savings

    The other speech to text solutions mediaTRUST had tried were up to 4x more expensive, and the manual transcription solutions were up to 10x more expensive compared to Vatis Tech. mediaTRUST has been able to reduce transcription costs.

  • Improved productivity

    1 hour of audio content was transcribed in less than 10 minutes, while manual transcription takes on average 3 hours. Vatis Tech speech to text software allowed mediaTRUST to monitor audio and video materials for its clients faster and be more efficient.

  • High quality content

    With Vatis Tech it was possible to improve certain keywords, primarily brands or products. A high content accuracy of over 95% could be achieved.

  • Business growth opportunities

    Ability to take on higher volumes of new business, based on media monitoring powered by speech to text AI solution, and thus fulfill even more complex projects with higher customer demands.

Next Steps

Move to automatic transcription of all media monitoring processes.

For engineers who read the docs before the marketing page

Read the documentation, try it free, tell us how it goes.