Resumen
The article reports that speech recognition technology has made dramatic progress over the past 30 years, triggered by advances in computing devices and architectures, success of algorithm developments, and availability of large quantities of data. Despite this progress, some fundamental and practical limitations in the technology have hindered its widespread use. There has been a very large performance gap between human and machine speech recognition. Mainstream adoption of speech recognition would not be possible without the underlying recognition technology that can deliver a sufficiently robust and low-error performance. Reducing the speech recognizer's error rate under all deployment environments remains one of the greatest challenges to making speech recognition more mainstream. Many leading researchers in the field understand the fragile nature of the current speech recognition system design.