Product Overview

ASR was built on the client/server paradigm (Fig. 4), with a layer providing the WebSocket and REST interfaces that allow the client application to access the speech recognition features. For phone applications with IVR, the standardized MRCP interface can also be used. To facilitate the development of applications, a few client libraries are provided that simplify the process of integration with ASR.

../_images/integration.png

Fig. 4 Integration with ASR

../_images/api_integration.png

Fig. 5 ASR Cloud integration.

To make application development easier, some client libraries are made available, which simplify the ASR integration process.

The acoustic model should be installed in ASR according to the language and the sample rate of the input audio.

The specific grammars for the application can be created by the developers themselves, with the help of tools provided by ASR. If the application uses the built-in grammars defined in the specification, they can be installed and used in the recognition. Veja mais sobre o uso de gramáticas em User Guide.

If the application needs to recognize spontaneous speech, a free speech model must be installed, according to the adopted language. The model generally used is the general context model, which can be used in any application. There can also be specific models for a given segment (medicine, sports, etc.), but they must be applied to the corresponding context.