About 12 % of the population suffer from speech and language disorders and about 1 % of people stutter. “Research has shown this influences their social, employment, academic and personal life in many negative ways,” says Dr Yair Shapira, founder and CEO of AmplioSpeech, the Israel-based company which leads the EU-funded ADST project. “Speech therapists do a very good job, but unfortunately, when it comes to stuttering, for every six children who go to therapy and succeed, another five go back to their old habit of stuttering – an 84 % relapse rate,” explains Dr Shapira, a biomedical engineer who started the company because his son stuttered. “We streamlined the whole speech-language therapy service using artificial intelligence,” he says. “With stuttering therapy using ADST, the failure rate is four times lower than without ADST. It’s a highly monitored, well-controlled process that results in much better outcomes.” Measuring speech efficiency The company is the first in the world to develop a system to measure the severity of speech-language impediments. “Once you can measure things you can start tracking progress,” Dr Shapira notes. “Speech is about communication, so we developed profound algorithms which detect efficient and inefficient speech and put a score on it. The normal range is 85-90 % efficiency. If your speech efficiency is, say, 50-55 %, it needs to be 30 % more efficient to be in the normal range.” Using a crowdsourcing platform, people were asked to rate a person’s speech too. “We found a very good correlation between what our systems and algorithms rate and what the general audience rates, which is really ground-breaking,” Dr Shapira says. The ADST system is based on what works in speech therapy, using some 20 million syllables and data points accumulated by the company. “That’s about 4 000 more pieces of data than that gathered throughout the research in speech pathology in the last 100 years,” Dr Shapira says, adding: “it’s not just high quantity, it’s also a high-quality database based on objective measurements. There was a lot of testing and collaboration with leading experts in different countries to validate the system,” Dr Shapira explains. Focus on practise “Speech therapy is not very structured, it is mostly intuitive,” Dr Shapira explains. In a clinical setting there is little tracking of progress and the completion rate for practice assignments is just 5 % when the child and therapist are not face-to-face. With the ADST structure, even less experienced speech pathologists can follow a thorough, yet flexible, therapy programme using good practice based on testing with thousands of people. Everything is done online, including regular sessions with the therapist, but with a lot of emphasis on practise assignments. A stream of tasks is made available to the child’s online application. “They get encouragement and notifications from our back office if they don’t do them. This is how we pushed the practice completion rate from 5 % to 86 %,” Dr Shapira says. Assignments can be monitored by the therapist, whereas in the past they often had little idea of what was happening between therapy sessions. “A child will say ‘of course I practised’,” Dr Shapira says. The company works with clinics in Israel, Portugal, France and the United States, having adapted the original English and Hebrew algorithms to work with other languages and is expanding the system to other speech-language disorders.
ADST, speech therapy, speech, language disorder, stuttering