Fabio Minazzi

What inspired the “Voice for Purpose” initiative at Translated?

At Translated, we like challenges, and we have a mission: enable everyone to understand and be understood in their own language. So when Italian actor Pino Insegno approached us with the challenge of using AI for the purpose of helping the voiceless, we recognized a new opportunity to accomplish our mission.

We already had many years of operating AI-based solutions for written communication under our belt. As a natural evolution, we had also developed high-quality voice algorithms for dubbing, with very satisfactory results. We studied the state of the art in speech synthesis for ALS patients and realized our infrastructure for running real-time interactions with linguists across the world could be a key element for delivering an innovative service. Basically, we started the “Voice for Purpose” project because we felt our experience and infrastructure could be used to help people.

What were the technical challenges of building a voice donation platform?

Voice donation is a new concept, and so is receiving a donated voice. Therefore, the first challenge was presenting those two new concepts in a simple way for people who land on the portal — so that they would feel confident in becoming part of the project, whether as a donor or a user.

Another challenge was making the platform open to everyone and still protecting it from inappropriate usage. Each step was designed and tested carefully, balancing simplicity and completeness, informality and security. Balancing the technical aspects with the user experience occupied much of our thoughts.

For example, we considered requiring the sample text that a donor records to strictly adhere to the script we provided. But then we realized that the majority of the donors are normal people who are not skilled in speech, and they might make mistakes; if we sanctioned deviations from script, we would have frustrated them and lost their precious contributions. We carefully measured the impact of those deviations from the script on the final result and decided that we would only ask the donor to retry if the sample was silent for any reason.

Now that we have collected the voices of more than 5,000 people of different ages, nationalities, and levels of education, we can say that keeping the platform simple was the right choice.

What aspect of the project are you most proud of?

There are several areas that make us proud. From the social point of view, we created one more opportunity for people to be generous — they can donate blood, cells, biological tissues, and now also voice. We have cases of people who recorded their voice for self-donation because they know they will lose their voice, which is also a new concept introduced by “Voice for Purpose.” These individuals are happy to make their voices available to other people, too.

From the medical point of view, we’ve discovered that donated voices have a measurable advantage compared with the standard, impersonal voices that are often used. Age, cadence, accent, and timbre can more easily be matched with a donated voice.

From the technical point of view, we are proud of having introduced real-time voice delivery for medical applications over the cloud, to scale — making the power of cloud computing available to any patient in the world. The algorithms we implemented are in fact too complex to be executed on many medical devices, which have not been designed with high-quality speech synthesis in mind.