NSC Releases Version 3.8 of Its Hardware-based Speech Recognition Engine

RISHON LEZION, Israel - Natural Speech Communication Ltd. (NSC) released version 3.8 of its hardware-based speech recognition engine.

The new version, an ASR Server running on NSC's family of PCI boards (NSCBoard™), introduces new product features and improved support for market standards and customer requirements.

The NSCBoard™ is a family of boards that includes two, four, 12 or 30 DSPs. The NSCBoard™-30 is a DSP-based board, designed to deliver up to 120 concurrent ASR channels in a PCI slot. Each DSP is autonomous and can be independently configured for a different language; for each ASR channel request, a different vocabulary and grammar can be used. Several ASR channels can run on one DSP, depending on the speech recognition task complexity. NSC has two new hardware developments:

• NSC has introduced a new board to its family of boards: NCSBoard2™ - to support small to medium business with up to eight ASR channels on one board.
• The NSCBoard™ family of boards was upgraded to support the PCIU standard.

The new version supports the ability of the engine to return multiple recognition results (N Best) to the application with a confidence level for each returned result.

Included in NSC's speech recognition engine, echo cancellation was implemented as part of the ASR resource. Echo cancellation is a needed feature to support "barge-in" capabilities - the ability to speak during the application voice prompts. The speech recognition engine includes an echo-cancellation capability, thus eliminating the need for having this feature on the telephony board.

Version 3.8 supports speech recognition in the following languages: North American English, UK English, Russian, German, Hebrew and Arabic.

The functionality of NSC's Server and APIs were enhanced to support VXML and MRCP specifications.

A Key Word Spotting (KWS) engine is included in version 3.8 with the capability to spot keywords spoken while monitoring on-going calls in real time or recorded calls.

The KWS engine is active for the whole call duration, looking for specific words within a long spontaneous and natural conversation between two or more speakers.

Using NSC's family of speech recognition engines, the NSCBoard™, the new KWS engine can provide real-time monitoring capability of thousands of simultaneous calls with system architecture.

Version 3.8 brings built-in grammars according to the VXML specifications. These are off-the-shelf grammars of standard and common phrases such as times, dates and natural numbers.

Version 3.8 supports ABNF grammar format. The new version also introduces support to Stochastic Grammars. Activation/de-activation of grammars is possible.

The NSC server is capable of handling any combination of board types and sizes while the same speech recognition engine runs on all boards. Version 3.8 introduces more advanced features such as:

• A configuration tool allowing autonomous definition of the NSCBoard™ speech recognition resources.
• New API commands that enable management of the NSC servers by the application. This means that the application can determine, in run time, whether to launch additional servers or shut-down running servers, according to the need.
• Can be operated on LINUX red-hat operating system. Overall - the NSC server supports Windows NT, Windows 2000, Windows 2003, Windows XP and LINUX.
• Defining recognition parameters.
• The ability for a run time configuration of the resources (remote/local) of the feeds to be recognized. The source feed for the NSC engine can be provided from a local host through physical direct connection to the telephony board or from a remote source via Ethernet.