Load-adjusted speech recognition

#The invention relates to speech recognition and more particularly to adaptive speech recognition with variable recognition computation.

#Computer-based speech-processing systems have become widely used for a variety of purposes. Some speech-processing systems provide Interactive Voice Response (IVR) between the system and a caller/user. Examples of applications performed by IVR systems include automated attendants for personnel directories, and customer service applications. Customer service applications may include systems for assisting a caller to obtain airline flight information or reservations, or stock quotes.

#IVR systems interact with users by playing prompts and listening for responses from users. The responses are attempted to be recognized and various actions can be performed in response to recognized speech.

#FIG. 1 is a simplified diagram of a user interacting through a telephone with a speech recognition system according to the invention.

#FIG. 2 is a simplified block diagram of a computer system shown in FIG. 1.

#FIG. 3 is a plot of usage over time of a processor of the computer system shown in FIG. 2 as a function of computational capacity of the processor.

#Embodiments of the invention provide techniques for adaptive speech recognition processing in both times of relatively light processing loads on a processor and relatively heavy processing loads. Relatively more processing capacity is used for speech recognition during times of relatively light loads and relatively less processing capacity is used for speech recognition during times of relatively heavy loads. The relative heaviness or lightness depends on the processing capacity. The amount of processing capacity used for speech recognition can be adapted/adjusted as a function of available processing capacity. For example, a load on a central processing unit (CPU) can be divided into categories of idle, normal, busy, and pegged. Using the amount of CPU capacity allocated for speech recognition during busy times as a baseline, the amount of CPU processing can be designated as 1.0 speech processing CPU unit. Exemplary CPU usages for speech recognition can be 2.0 CPU units during idle times, 1.5 CPU units during normal-load times, 1.0 CPU unit during busy times, and 0.8 CPU units during pegged times.

#Referring to FIG. 1, a user 2 is connected through a telephone 4 to a speech recognition system 10. The user 2 can interact with the system 10 through the phone 4 in a normal manner over a line 51. As shown, more lines 52-540 for a total of 40 lines, are provided for other users (not shown) to interact with the system 10. The system 10 includes an Interactive Voice Response (IVR) system 6, that includes a Central Processing Unit (CPU) usage controller 8. The controller 8 is configured to determine amounts of processor capacity that are available for speech recognition and to adapt recognition techniques in accordance with processor availability, as described more fully below. The system 10 is implemented as a computer system as shown in FIG. 2.

#Referring also to FIG. 2, the speech recognition system 10 includes a processor 12, memory 14, disk drives 16, a display 18, a keyboard 20, and a mouse 22. The processor 12 can be a personal computer central processing unit (CPU) such as a Pentium® III processor made by Intel® Corporation. Memory 14 includes random access memory (RAM) and read-only memory (ROM). The disk drives 16 include a hard-disk drive and can include floppy-disk drives, a CD-ROM drive, and/or a zip drive. The display 18 is a cathode-ray tube (CRT), although other forms of displays are acceptable, e.g., liquid-crystal displays (LCD) including TFT displays. The keyboard 20 and mouse 22 provide data input mechanisms for an operator (not shown) of the system 10. The components 12, 14, 16, 18, 20, and 22 are connected by a bus 24. The computer system 10 can store, e.g., in memory 14, software code containing instructions for controlling the processor 12 to perform functions described herein. Due to the nature of software, features and functions described herein may be implemented using software, hardware, firmware, hardwiring, or combinations of any of these, that may be physically located other than as described, including being distributed in various locations.