Microchip Technology announces dsPIC30F speech recognition library

EDP Weekly's IT Monitor, March 28, 2005

Microchip Technology Inc. (NASDAQ:MCHP), a provider of microcontroller and analog semiconductors, today announced a speech recognition library for its 16-bit dsPIC Digital Signal Controllers (DSCs). The dsPIC30F Speech Recognition Library enables embedded designers to concentrate on their primary application while adding valuable speech recognition features in the same way they would add a soft peripheral. This library provides engineers with a way to use one dsPIC DSC for both controlling the application and recognizing speech commands, without being an expert in speech recognition.

Many design engineers are seeking to add speech recognition to their designs in order to further differentiate product offerings and provide users with a natural, less-complex user interface--particularly in environments where hands-free operation is important. The dsPIC30F Speech Recognition Library, in tandem with the dsPIC DSC, is especially attractive for designers who have space constraints or wish to cost-effectively add speech-recognition functionality. With its integration of a 16-bit microcontroller and a DSP in a single architecture, the dsPIC30F can accommodate primary control applications while performing speech recognition using excess memory and MIPS.

"Many of our software libraries conform to standards that make their efficacy quite predictable. Speech recognition differs from these libraries, in that the apparent accuracy is in the eye of the beholder," said Sumit Mitra, vice president of Microchip's Digital Signal Controller division. "With that in mind, we allow our full dsPIC30F Speech Recognition Library to be test driven for $5 so our customers can make their own determination of suitability."

Any embedded application with speech-based input can benefit from this advanced speech recognition library. Specific example applications include: hands-free cell phone kits, emergency monitors for the elderly, appliances, portable music players, toys, home security systems and programmable electronic products such as VCRs, DVDs and thermostats.

The dsPIC30F Speech Recognition Library provides isolated, speaker-independent word recognition of North American English. Through isolated word recognition, the software allows a user to control an application via a set of fixed voice commands. A 100-word library has already been pre-trained by a demographic cross-section of male and female U.S. English speakers to maximize recognition accuracy. A PC-based Word Library Builder utility allows a custom library subset to be selected from the full 100-word library (for a given application). And, because it is speaker independent, no training is required for end users of the product.

The speech recognition library requires 9 MIPS, which leaves significant performance headroom for the main application in a dsPIC30F with 30 MIPS. The library is designed for dsPIC DSCs with a minimum of 33 Kbytes of Flash and 4 Kbytes of RAM, along with an integrated codec interface. Today, those products are the dsPIC30F5011, dsPIC30F5013, dsPIC30F6012 and dsPIC30F6014, with additional DSCs planned. The library and PC-based word selection builder are described in detail in the dsPIC30F Speech Recognition Library User's Guide (DS70140) and Word Library Builder User's Guide (DS70137), respectively. Both guides are available on Microchip's Web site.

The speech recognition system can be evaluated on a dsPICDEM 1.1 General Purpose Development Board (DM300014). This board supports both the full speech recognition library and a fixed, 10-word speech recognition demo that can be downloaded for free from Microchip's Web site.