Superior performance with HyperVox™ acceleration
Our proprietary acceleration technology makes large-scale speech recognition affordable, allowing text and speech analytics providers to serve more customers with broader speech coverage.
The Voci™ appliance is engineered for extremely fast, accurate, and comprehensive speech recognition through an application-specific architecture implemented with a commercial off the shelf hybrid-core computer. The system is a combination of a server-grade Intel® Xeon® processor and Voci’s breakthrough HyperVox hardware accelerator technology, bringing together the best of two worlds —the flexibility of an enterprise server with the high performance of custom hardware. With Voci’s revolutionary server appliance, speech recognition is finally liberated from the constraints of software-only solutions. Voci’s technology provides unparalleled advantages, including:
Hybrid-core computers are the platform of choice for many high performance applications, combining an industry-proven general purpose processor with high throughput, application-specific compute engines. The Voci™ hybrid-core computer runs a standard enterprise edition Linux OS on an Intel Xeon processor, interconnect and I/O chipset, and employs state-of-the-art configurable Xilinx® Field Programmable Gate Arrays (FPGAs) to power our HyperVox™ speech accelerator. The appliance is packaged in a standard 2U rack-mountable form factor, and is configured to be easily upgradable. Enhancements to the Voci HyperVox engine appear as software upgrades and require no physical hardware changes. The result? Network administrators experience a standard Linux server, while users experience significant improvements in performance - and managers see dramatic cost savings.
Building on over 10 years of research, the Voci™ HyperVox™ technology accelerates speech recognition by orders of magnitude over comparable commercial solutions. Unlike conventional speech recognizers which are restricted to general purpose processors, Voci’s speech recognition-specific architecture pushes the performance envelope by dedicating the essential resources to each aspect of the speech recognition algorithm. Parallel functional units speed up compute-bound regions, while custom high bandwidth/low latency memories speed up memory-bound regions. By focusing on optimizing speech recognition as a whole, HyperVox achieves levels of performance that are unattainable with other hardware acceleration platforms such as Graphics Processing Units (GPUs) or Digital Signal Processors (DSPs). HyperVox is designed to be extremely flexible to support speech applications of any size, and accelerates recognition across any domain and configuration. This is achieved because only speed-critical components are fixed into hardware, allowing speech parameters to remain configurable in software form. The technology supports both large vocabulary and phonetic recognition. Using Voci’s easy-to-use API, HyperVox can be adapted to any speech solution to provide a significant boost in performance.