Personalizing Speech Output for Communication Devices

描述（由申请人提供）：ModelTalker是一个文本到语音合成（TTS）系统，旨在对潜在的增强和替代通信（AAC）设备用户特别感兴趣。目前大多数AAC设备使用的TTS软件源自20世纪70年代开发的基于规则的形成峰合成算法，但现代串联合成系统（ModelTalker就是一个例子）可以提供更自然的合成语音，具有同等或更高的可理解性。此外，虽然基于规则的形成峰合成通常产生的语音不像特定的说话者，但串联合成提供了捕获特定个体声音的可能性，因为它是基于对单个说话者的语音进行采样。这种产生个性化，独特的合成声音的能力是AAC设备用户连接合成的最大吸引力之一。该STTR申请寻求资金，将ModelTalker TTS技术从研究实验室转移到小型企业，以实现商业化。在转换过程中，提出了对合成软件的具体增强，并将开发一种新的语音，以(a)证明ModelTalker TTS系统在自然性和可理解性方面与现有TTS系统具有竞争力，以及(b)提供将随系统商业版本一起发布的几种默认语音之一。该项目的长期目标是通过将ModelTalker引擎与当前正在开发的语音创建工具打包，将个性化合成“声音”的方法带给一般的AAC人群。预计这项技术将在人们想要独特的合成声音的应用中具有更广泛的吸引力（例如教育软件和电子邮件听写）。

DESCRIPTION (provided by applicant): ModelTalker is a text-to-speech synthesis (TTS) system intended to be of particular interest to potential Augmentative and Alternative Communication (AAC) device users. Most present AAC devices use TTS software that derives from rule-based formant synthesis algorithms developed in the 1970s, but modern concatenative synthesis systems (of which ModelTalker is an example) can provide much more natural sounding synthetic speech of equal or higher intelligibility. Additionally, while rule-based formant synthesis typically produces speech resembling no particular speaker, concatenative synthesis affords the possibility of capturing the voice of a specific individual because it is based on sampling the speech of an individual speaker. This capability to produce personalized, unique, synthetic voices is one of the strongest attractions of concatenative synthesis for AAC device users. This STTR application seeks funding to transfer the ModelTalker TTS technology from the research laboratory in which it was developed to a small business for the purpose of commercialization. In the transfer process, specific enhancements to the synthesis software are proposed, and a new voice will be developed to (a) demonstrate that the ModelTalker TTS system is competitive with existing TTS systems in terms of both naturalness and intelligibility, and (b) provide one of several default voices that will ship with a commercial version of the system. Long Term goals for this project are to bring the method for personalizing synthetic "voices" to the general AAC population by packaging the ModelTalker engine with voice creation tools currently being developed. It is anticipated that this technology will have broader appeal for applications in which people want a unique synthetic voice (e.g. educational software and email dictation).