Design

Designing a speech response module for a given application is always a challenging problem. The machine itself represents a certain platform. Its mobility and communication capabilities drive design considerations, as does the economics of the enterprise. Low cost, high effect solutions constitute the best designs traditionally, but sometimes how the short term tradeoffs offset the long term tradeoffs becomes a very important factor in the design. Other significant considerations that go into designing speech-enables applications include the range of environments that can and are expected to be handled by the machine, and its noise exposure. It is well-known that high noise environments can degrade speech response systems significantly. Additional modules need to be designed when a machine is expected to operate in high noise and critical environments.

Accuracy of recognition is another important factor. In recognizing a credit card number, it is necessary to achieve a 100% accuracy: an all-or-nothing scenario. In a web-browsing application, just recognizing some some keywords accurately may suffice: a something-is-better-than-nothing scenario. A speech response system can be controlled through external knowledge sources to make it respond in one of such many possible scenarios.

Human factors

Designing a speech response module for a given application is always a challenging problem. If the machine is not complicated, the human being who uses it, certainly is. People not only speak differently from each other, they speak differently at different times, and the gamut of emotions that an average human being runs through on any average day reflects very clearly in their speech. In addition, a person may speak many languages, and when the spoken commands are uttered in a non-native style, their content becomes difficut to understand. Current technology has a hard time dealing with this wide range of human variations.

page 4