Is Speech Recognition Dead?

In his recent blog post, “Rest in Peas: The Unrecognized Death of Speech Recognition,” Robert Fortner posits that the accuracy of speech recognition software peaked in 2001 at 80% and, as of 2006, had still not progressed beyond that level. In contrast, a human can transcribe at an accuracy rate as high as 98%.

Many pundits in the speech recognition industry argue that their software currently has a 98% accuracy rate. If you take a closer look at those numbers, you will find that these accuracy rates are based on learned responses similar to those used in automated attendants. In over three decades of research, no parser of unrestricted text has been developed. No machine has established a fundamental understanding of our spoken language.

The bottom line is: to ensure that your speech is accurately transcribed, you need a human. Only humans can understand the nuances of the spoken language. Only humans can ensure that your transcription of conversational speech is 99% accurate or better. And, as an added bonus, you do not have to train a human to understand your voice – just record your speech and give the audio file to your transcription company.

May speech recognition software rest in peace.