Junctures in speech communication

Abstract

Several junctures in research on speech communication seem salient over the past 50 years. Under the goad of expensive long-distance transmission, and the criticality of voice communication in a world in turmoil, bandwidth compression methods received strong emphasis in the 1940s. Because transoceanic communication was by HF radio, privacy and encryption were also central issues. These incentives resulted in the Channel Vocoder and a variety of offshoots. By the 1960s, digital computers were being deployed widely and sampled-data theory was well established, providing the systems researcher with a new environment for experimentation—one in which concepts could be rapidly simulated and evaluated. Bandwidth compression continued to be a focus, as computer simulation techniques progressed from classical filter emulation to new designs based solely on digital theory. By the 1980s, broadband connectivity (including transoceanic cable) was becoming pervasive and economical enough that emphasis on bandwidth conservation waned. Concomitantly, digital machines were expanding in capability—but were markedly limited by their inability to communicate with human users in natural ways. The focus consequently shifted to human/machine communication—especially by conversational means. As we approach the year 2000, a strong research emphasis remains on human/machine communication—but with expanded aspirations for natural interaction. Computer interfaces with multiple modalities—utilizing sight, sound, and touch in combination—are beginning to serve multiple collaborating users. In the same era, explosive deployment of two technologies—cellular telephony and computer networking—has rejuvenated interest in bandwidth conservation and coding for privacy—with many of the early vocoder concepts evolving to sophisticated digital forms implemented on single-chip processors. As we move toward 2020, a special emphasis is likely to be on multilingual communication. And in this time, continued advances in computing, acoustic signal processing, and language modeling should carry translating telephony far beyond the ‘‘phrase book’’ stage.