Design example of voice application solution for handheld devices based on TI's OMAP platform

The OMAP platform provides a perfect solution for developing voice applications for personal handheld devices. This low-power OMAP architecture combines DSP signal processing functions for voice with the general system performance of a RISC processor. An open software architecture is designed to encourage the development of complementary applications such as voice engines, voice applications and multimedia. Development support, including voice recognizers and prototype applications, can help developers quickly build their own products and shorten time to market. The OMAP platform ensures that developers can seize the growth opportunities of personal handheld devices by easily and flexibly adding voice applications. The

application of voice technology is increasing, bringing rare opportunities for application developers to add high-value functions to handheld devices, mobile devices and wireless personal devices. Today's personal handheld device voice is mostly limited to voice dialing, but technologies have emerged for the wider development of voice recognition and text-to-speech applications. Developers who intend to add voice functions need to be familiar with all aspects of voice technology. These issues include not only processing and memory requirements, but also how specific platform architecture and support can facilitate the development process and shorten time to market. Adding value with voice applications can  

bring rich potential benefits. According to estimates from various market research firms, the compound annual growth rate for personal handheld devices is expected to reach 20% over the next two years, with total global device shipments reaching 700 million units by 2004. To tap into this huge market with value-added voice applications, developers must turn to underlying technologies that give them high performance, low power consumption, and support that helps them quickly launch new products.  

Voice capabilities provide users with a natural way to input and output, and are safer than other forms of I/O, especially when the user is driving. In most applications, voice is an ideal complement to keyboards and displays, not a replacement for them. For example, in very noisy environments, listening and speaking may not be practical, so users may have to rely on keyboard input and display reading. Similarly, users often prefer to enter certain things, such as PIN numbers and passwords, on the keyboard rather than speaking them out loud for others to hear.  

Voice dialing is the most commonly used voice technology in today's personal wireless devices. Voice dialing usually allows calls without the need for hands and ears, which is a particularly important feature when driving. Voice dialing includes name dialing, which is to call a name in the address book, and number dialing, which is to speak a phone number. As shown in Figure 1, other potential voice applications include:  

1. Voice email?D?D Includes browsing mailboxes, writing emails using voice input, and listening to emails read out.  
2. Information retrieval?D?D Stock prices, headlines, flight information, weather forecasts, etc. can all be retrieved from the Internet via voice. For example, instead of first going to a website and typing in a stock name or browsing a predefined list, a user can command: "My stock quote, Texas Instruments."  
3. Personal information management?D?D Allows users to specify appointments, view calendars, add contact information, etc. by voice.  
4. Voice browsing?D?D Using a voice program menu, users can surf the web, add voice favorites, and listen to the web page content read out.  
5. Voice navigation?D?D A full voice input/output driving system for obtaining navigation under automatic and eyes-not-enough conditions.  

  
Speech Technology Issues  

Speech systems must meet certain basic requirements for use. Obviously, the speech output must be clear and understandable to the user. ASR must also support natural speech for a given application use. What is natural can vary widely, ranging from simple names and commands spoken word for word to continuous sentences with large vocabulary. In addition, people's natural speech and pronunciation patterns vary, so the system should be flexible to accept different speakers. The recognition engine must be accurate, otherwise users will not use the technology. The  

system requirements for speech are processing intensive and may involve large amounts of memory, depending on the vocabulary supported. For server-based applications, wireless bandwidth usage will increase. These factors will also affect other system considerations. The higher the MIPS and transmission requirements of the application, the higher the power consumption of a given system, thereby reducing battery life or requiring more frequent charging. Response time may also increase when the application requires the use of processor external memory.  

Certain application trade-offs can help reduce system requirements by abandoning unnecessary functions for handheld devices. A speaker-based system that recognizes only a small number of words and discrete speech will require significantly fewer resources than a speaker-based system that recognizes a large vocabulary and continuous speech. Support for additional languages ​​increases processing requirements and doubles the memory required for an application. Noise and interference immunity are important features, but increase complexity and memory requirements. [page]

Obviously, developers want to minimize the performance degradation of the basic application when adding features such as speaker dependence, continuous speech, vocabulary size, and language support. There are options that help reduce the performance degradation in speech technology, such as distributed speech recognition (DSR). DSR splits the recognition task so that the handheld device can convert the raw speech into spectral feature vectors while the server performs the recognition process. This approach and similar distributed TTS approaches rely on standardization of processing methods and transmission protocols. Although these technologies are promising, developers are still faced with limited resources for speech applications in personal handheld devices.  

Therefore, choosing the right platform for high-performance applications such as speech is as important as carefully designing the application's functions. Such a platform must have strong processing power while achieving a high level of power efficiency, not only in core operations but also in processing memory. There should be enough MIPS to support multimedia, security, and other complementary applications. It is also important to provide programmability to integrate new algorithm capabilities. Finally, the platform must include a software architecture designed to support modular application development to help developers quickly bring products to market.  

OMAP Technology: An Excellent Voice Platform  

TI's OMAP platform provides an excellent solution for developing voice applications in personal handheld devices. The dual-core architecture of the OMAP1510 and OMAP5910 processors integrates the power-efficient TMS320C55x? digital signal processor (DSP) and the high-performance ARM9RISC microprocessor. As a result, these OMAP processors provide the arithmetic-intensive signal processing capabilities required for voice while also providing the general-purpose performance required for system-level operations. The OMAP710 processor is a highly integrated single-chip solution with a DSP-based GSM/GPRS baseband for wireless communication processing and a dedicated TI-enhanced ARM925 processor for low-power execution of multimedia applications. The OMAP1510, OMAP5910 and OMAP710 processors support low-end ARM-based voice applications. They also have code compatibility, allowing developers to integrate software applications into individual products for different markets. The OMAP1510 and OMAP5910 feature DSP processing capabilities for more intensive voice applications.