From Speech Strategy News, October 2007

Sensory adds speaker-independent speech recognition for CSR Bluetooth chips

Hands-free headsets and car kits can have embedded voice control

CSR provides personal wireless technology, including a Bluetooth chip solution (the BlueCore5-Multimedia platform) that is widely used in wireless headsets and hands-free car kits. CSR customers include Apple, Dell, LG, Motorola, NEC, Nokia, Panasonic, RIM, Samsung, Sharp, Sony, TomTom, and Toshiba.

On September 18, Sensory Inc. announced that it had ported its embedded speech recognition and speech synthesis solution to the BlueCore5-Multimedia platform, and that it was available through CSR’s eXtension Partner Program. The FluentSoft Bluetooth Suite from Sensory will allow CSR’s customers to add speaker-independent voice control as software to Bluetooth products using the chip set. Fitting speaker-independent speech recognition into a chip that itself can fit into a battery-powered earpiece is a significant challenge, notes Todd Mozer, Sensory president and CEO.

Sensory software can significantly improve the usability of the devices, Mozer said. Since a Bluetooth hands-free microphone can have minimal controls (typically one button), that button must normally suffice for all commands. For example, one function might involve one quick press, another two quick presses, and a third a long press. Not only is such an interface clumsy, it is severely limited. Speech recognition can have an indefinite number of commands; a button can act as a push-to-talk switch. There could be voice commands for call control and dialing, battery level and connection checks, pairing (linking to Bluetooth devices), adjusting audio settings, and playback and control of music.

The FluentSoft Bluetooth Suite in this application differs from typical embedded solutions in wireless phones. Relative to the BlueCore5 used in a headset, wireless phones have a faster processor, more memory, and a bigger battery to support that hardware. They can use the contact list in the phone, for example, with software to convert the text to a phonetic representation, and then recognize any name in the wireless phone. There is no Bluetooth protocol for retrieving the contact list from a wireless phone, even if there were enough processing power and memory to create a dynamic list, Mozer said. Instead, developers using the software must decide ahead of time what speaker-independent vocabulary they want and load it with the software.

Digits are an obvious part of the vocabulary, to allow dialing by speaking a series of numbers continuously. Other commands  that would otherwise require a button press (such as “hang up”) can be included. “Speed dial 1, speed dial 2,…” could be used to dial presets if the user could remember what the presets correspond to. In a particularly innovative feature, Sensory has provided “Goog-411” as a vocabulary word that dials the Google local voice search number 1-800-GOOG-411. Thus, if a developer uses that speaker-independent vocabulary word, the feature allows quick connection to a network-based voice portal. Headsets and car kits distributed with this feature could add millions of users for such services—Google or otherwise. FluentSoft supports all primary world languages, and enables OEMs to easily customize the user interface, command vocabulary, and voice prompts to create a unique, differentiated experience for their customers.

Sensory’s SX speech synthesis is also included. SX is speech-compression software, not text-to-speech software. High quality voice prompts can be recorded for guidance and for providing answers to status queries such as battery check and guidance on device operations. An incoming call can be announced—“Call from 408-625-3300, would you like to accept?” The result is a much more satisfactory user interface than provided by the beeps and other cryptic audio signals of current solutions, Mozer emphasized.

Anthony Murray, Senior Vice President of CSR’s Wireless Audio Strategic Business Unit commented, “Bluetooth mono and stereo headsets are decreasing in size, and speech recognition overcomes the challenges faced by a more limited user interface…we can now provide a single-chip solution with Bluetooth technology and speech recognition, considerably reducing manufacturing costs of hands-free car kits and embedded Bluetooth designs.”

Sensory chip and software speech solutions have shipped in more than 50 million units for hundreds of customers, Mozer said.