Multilingual Machine Voice with a Human Touch

Intento Machine Voice

Last year, Konstantin Savenkov (Co-founder and CEO, Intento) and Carrie Fischer (Manager of Globalization Services, Subway) came together to present a case study on how MT and speech technologies are already delivering very real results in scaling multilingual content creation in global launch cycles. We’ve been seeing a surge of interest in machine voice, and wanted to share the case study for those who want to learn more about potential use cases for their own business in e-learning and beyond. 

The talk touched on one aspect of the work that Intento does with Subway to boost their ROI — translating training videos. Here, we’ll focus more on how exactly this solution helps Subway’s ROI rather than the nitty-gritty of how it works. We’ll go back to the basics and investigate the returns that Subway saw during this process.

Learning to work with MT  

For years, Subway had been hesitant to begin experimenting with Machine Translation. This is often the case for companies that simply haven’t been exposed to the benefits of working with MT. Without real-life experience, it’s easy to look at the MT landscape and feel overwhelmed with the array of options and be skeptical about the results. You sit there and think, ‘it’s too weird, too creative; it’s not for me’.

As with many localization managers, Carrie reached a point where she felt swamped with the amount of content she needed to have translated and the limitations of their established system. She and her team decided that even though they didn’t have much MT experience, it was time to listen to the experts who have insights on the entire MT landscape and the tools to create and translate content 4 times faster with 97% user satisfaction.

Intento began by focusing on MT for Subway — everything from knowledge-based articles to the franchisee website. However, what surprised them the most were the impressive results working with voice synthesis and e-learning content.

Subway’s e-Learning Process 

During the height of the COVID-19 pandemic, Subway had to close all of its global learning centers. This is where franchisees would go for two weeks to learn how to open and run their restaurants. Closing all of the live courses meant mass digitization of content to e-learning modules. It meant having eight distinct characters interacting through speech bubbles and text on screen with 8 distinct voices. That all boiled down to 2.1 million words in 10 target languages with voice-over — ASAP.

Savings: Translation and Voice-over 

Just looking at the four big volume languages (French, German, Japanese, and Spanish), the total word count was ~1.5M.

Focusing on the percentages for the translation alone — Subway was able to save 46%. Savings in time were as significant as at least two months of human labor to come up with translations of this level of accuracy. For the voice-over, Subway was able to save 41% and about one month of human labor.

The ROI takes into account all three key initiatives in which Subway used MT and voice synthesis. By doing this, they saw in one year a 1,000% return on their investment. This doesn’t even factor in the unrealized ROI for less quantitative statistics. For example, Subway was able to open more new restaurants than they typically do once they efficiently brought the training to a virtual platform.

The Main Factors Increasing ROI 

The dramatic increase in ROI is realized through a few channels.

  • Intento is automating what is already there (existing workflows, software, and LSP vendors).
  • We have very low upfront costs (in the thousands) and a fast time-to-market turnaround (weeks).
  • Output quality (same as you’re used to getting with human labor)
  • Quick feedback in terms of realizing ROI statistics (weeks)

Especially looking at this last point, a solid ROI framework helps build the connection between localization initiatives and executive directors. It’s a simple way to prove to your company what MT is capable of achieving without waiting years to show off your innovative approach.

For localization, your ROI can be simplified in the following way, as it was in the case of Subway:

The Main factors Increasing ROI

If you found this interesting, Konstantin Savenkov will be speaking on ‘the GPT-3 disruption’ at SlatorCon Remote March 2023. Please join us to learn about how text generators will influence the machine speech and MT landscapes. Now’s the perfect opportunity to hear it from the experts!

Intento will also be releasing a chapter of the Intento Machine Translation University on GPT soon. Subscribe here to get notified upon its release. In the meantime, you can read about solving your most pressing business problems with AI, leveraging the entire MT landscape for your benefit, and more.