Spoke : a framework for building speech-enabled websites

[摘要] In this thesis I describe the design and implementation of Spoke, a JavaScript framework for building interactive speech-enabled web applications. This project was motivated by the need for a consolidated framework for integrating custom speech technologies into website backends to demonstrate their power. Spoke achieves this by providing a Node.js server-side library with a set of modules that interface with a handful of custom speech technologies that can perform speech recognition, forced alignment, and mispronunciation detection. In addition, Spoke provides a client-side framework enabling some audio processing in the browser and streaming of the user;;s audio to a server for recording and backend processing with the aforementioned integrated speech technologies. Spoke;;s client-side and server-side modules can be used in conjunction to create interactive websites for speech recognition, audio collection, and second language learning, as demonstrated in three sample applications. Though Spoke was designed with the needs of the MIT Spoken Language Systems group in mind, it could easily be adopted by other researchers and developers hoping to incorporate their own speech technologies into functional websites.

[发布机构] Massachusetts Institute of Technology