Siri & Co. often do not understand stutterers properly
However, human language is extremely diverse and rich in nuances in many ways. This often causes problems for Siri & Co., for example with strong dialects or restrictions due to illness. People with speech or speech disorders are particularly hard hit by the difficulties in understanding digital assistants. For example, stutterers are often faced with Siri mistaking their breaks for the end of their command, which leads to suboptimal results. In these cases, the request can usually not be answered correctly.
“Hold to Talk” doesn’t always make sense
On the iPhone and iPad, the problem can be mitigated by holding down the home button or the side button. This feature, known as “Hold to Talk”, is hardly useful on other devices such as the HomePod and also contradicts their operating concept. For some time now, Apple – like Google and Amazon – has been looking for ways to teach the in-house language assistant a better understanding of people with language or speech disorders. The reported the Wall Street Journal.
Siri should automatically detect stuttering
Apple’s goal: Siri should automatically recognize that a speaker is stuttering and then deal with breaks much more tolerantly. As part of a research project, the iPhone group has therefore compiled a collection of around 28,000 corresponding audio clips from various podcasts. These contain five different forms of stuttering and serve as the basis for training the system. Apple has now published the first results in a paper (PDF file). Accordingly, speech understanding improved by around 28 percent in the cases examined. The company assumes that this can be significantly improved by combining it with other measures.
Investigation of further language and speech disorders
In the current research project, Apple is exclusively concerned with the automatic detection of stuttering. The iPhone company expressly points out, however, that other language and speech disorders such as dysarthria could also be included in the further course of the work. It is not known when the results of the study will flow into Siri and give Apple’s voice assistants increased tolerance.