Veridict SpeechCAD

Veridict SpeechCAD™ is an integrated environment for expert customization and modeling of your speech-enabled system. It gives the user a powerful and flexible platform for customization and fine-tuning of speech driven applications built around the Daytona speech engine.

With SpeechCAD users can easily model every aspect of the speech engine’s behavior. It features easy-to-use functionality for all relevant aspects, including language modeling, acoustic modeling and adaptation, user profiles and management, pronunciation modeling and integration of custom words. The software environment also includes tools for testing and verifying speech recognition performance, running simulations both online and offline and measuring accuracy.

Using SpeechCAD, the user can calibrate the recognizer’s capabilities using an easy-to-use graphical interface and build customized profiles for individual users or other specific requirements. Typically, profiles are adapted towards one or several of the following criteria:

• Speaker - Voice quality, speaking style, accent and pronunciation.
• Language domain - Custom or application-specific vocabularies and language
• Environment and channel - Acoustic environment, ambient noise, microphones, etc.

Refined speaker-adapted acoustic models are easily created in SpeechCAD through data-driven processes and automatic learning by supplying additional speech data. This can be performed either by loading prerecorded speech data or by letting users speak directly into the SpeechCAD modeling tool. Support and conversion tools are provided for all common file formats. The adaptation process utilizes existing generic acoustic models, which have been trained on several thousands of speakers of the same language. By adding the speaker-dependent data on top of a generic model, optimally performing speaker profiles can be obtained by only adding a small data set.  Generic speaker-independent models are included, for both male and female speakers. 

Veridict speech technology provides support for, and has been tested with, a wide range of microphones and headsets on the market, including audio-capture devices from all larger vendors, including Andrea, AKG, Sennheiser, Shure and VXI.  Custom profiles are available for non-standard audio-capture devices, such as ear-canal microphones. Support for additional non-standard microphones can be added easily through Veridict’s modular software architecture.