Speech Recognition Hardware Solution - Hardware ArchitectureNSC's unique and innovative hardware product offering provides a server-based, multi-engine, speech product that is deployed on a dedicated PCI blade. The NSC SpeechBlade™ platform supports the required memory and processing resources for the speech recognition task. It enables full scalability from incorporation of speech recognition into an existing application server (“all-in-one-box” solution) up to addition of multiple servers allowing for deployments of a large number of speech recognition channels. Key Features in NSC's HW Product offering :
The Concept - Speech Server on a Blade
NSC provides multiple speech processing engines operating on the NSC SpeechBlade™ product line - PCI blades comprising a unique platform that enables the entire speech processing operation to be done on hardware. This unique approach by NSC allows a "CPU-free", high density product with simple system architecture. NSC’s hardware-based products implement NSC's unique technology through three high-level components:
The Engines
The Platform – SpeechBlade™
Language Packages NSC’s products run on Windows (XP, server 2003 R2-32Bit and Win7 ) and Linux and support multiple local and remote applications. NSC Hardware-Based Speech Recognition EnginesThe speech recognition engine part of the hardware-based product includes all the product software components. The speech engine consists of firmware that operates on the NSC SpeechBlade™ , NSC’s PCI blade, as well as server software, application development tools, and APIs. The NSC SpeechBlade™ Server includes two main speech engines for the hardware platform that implement the company's unique technology and operate on the NSC SpeechBlade™ product line of PCI blades. These two speech engines deliver different speech recognition functionality on NSC’s hardware platform and target the needs of different market segments.
NSC Spotter
NSC Speecher NSC Hardware-based Speech Recognition Engine Components:
NSC SpeechBladeThe Platform – The NSC SpeechBlade The NSC SpeechBlade™ is NSC's high-density, cost-effective PCI blade product line comprising a unique platform that supports running multiple telephony speech processing engines. The blades are composed of basic modules, each including a DSP, memory and peripherals. Each module is autonomous and can be independently configured to a different speech engine, language, and resource type. Several speech processing resources can run on a single module depending on the speech processing task complexity. The entire speech processing operation is done on the PCI blade – thus creating a robust, CPU-free platform for the operation of multiple speech processing engines. The product line of blades allows for full scalability from very small capacity (e.g., 2 speech recognition channels) to very high capacity (e.g., 90’s of speech recognition channels in a single server):
The NSC SpeechBlade™ has a PCI interface for communication with the NSCServer and an H.100 bus interface for acquisition of speech samples directly from a Computer Telephony Integration (CTI) board. The Speech Acquisition ProcessEach speech resource acquires speech samples independently. There are three alternative interfaces for acquisition of the speech samples:
Through the PCI Interface, API commands can be issued and language packages and firmware can be downloaded.
NSC SpeechBlades™ offer Optimum Flexibility NSC SpotterNSC Spotter - A Robust Real-Time Word Spotting Engine for NSC’s hardware platform driving Audio Surveillance, Speech Analytics, and data mining applications NSC Spotter is the Keyword Spotting (KWS) engine for NSC’s hardware platform designed for locating words out of a pre-defined list in spoken conversations. It operates in real time or on recorded calls, and is targeted for the Security and Intelligence and Call Center markets. NSC Spotter is used for speech analytics and call surveillance in the Security and Intelligence market as well as for mining business information and monitoring agent performance in the Call Center market. In addition, NSC Spotter can be used in any market segment, service or application where there is a need for analysis of a large number of speech calls or recordings. NSC Spotter implements a unique keyword spotting algorithm - a one-stage approach. The approach searches for keywords in a one-stage operation. The recognition is phoneme-based - NSC Spotter looks for the complete word in the speech stream based on a target sequence of phonemes that represent the keyword. The unique approach by NSC that operates directly on the incoming speech in one stage rather than the usual two-stage system that translates the speech into text (phonemes or words) and then performs a text search for the keyword. NSC Spotter has the following main features:
Read the KWS for Speech Analytics White Paper NSC SpeecherNSC Speecher - A High-Accuracy Speech Recognition Engine for NSC’s hardware platform targeted at Telephony Speech-Driven Services NSC Speecher is the telephony speech recognition engine for NSC’s hardware platform designed for speech enabling services in the Call Center, Telco, and Enterprise markets. Its main focus is IVR (Interactive Voice Response) applications. NaturalASR speech enables services such as: Call center automation, information access, automated attendant, value-added services, call routing, and voice controlled voicemail. NSC Speecher has the following main features:
|
