N/V

TaskSpeech® Plus YesNo (TS+ YesNo) is a standalone speech and voice laboratory tool for prototyping yes-no dialogues. It uses text-to-speech (TTS) synthesis to ask questions, and automatic speech recognition (ASR) to capture a user’s answers. By giving voice user interface designers (VUIDs) direct hands-on experience, it promotes a deeper understanding of conversational design, including error-recovery and turn-taking. The purpose of the app is to teach prompt-response design principles, and to support development of yes-no interactions for functional and usability testing.
TS+ YesNo is perhaps the most basic of TaskSpeech conversational modules. It is often used as the simplest information-extraction event in a dialogue. It may also be embedded into more complex SpeechBehaviors™, supporting confirmation and other forms of grounding.
Yes-no questions can be surprisingly subtle and varied. Each event begins when the app asks a well-phrased question. When the app has finished speaking (prompting), the user must reply to the question. The reply might be a simple “yes” or “no,” might present a synonymous phrase, or might even contain unrelated out-of-vocabulary (OOV) utterances. This give-and-take might happen repeatedly, exhibiting a prompt-response dialogue style that can be applied to highly interactive exchanges.
TS+ YesNo provides both educational and speech-research value in a small-footprint package, and is for designers, programmers, researchers, teachers and learners. The resulting knowledge informs the design, development and testing of any speech and voice applications you may be considering. General features include:
- Comes with default vocabulary and settings -- just download, tap and speak.
- Create your own vocabulary and settings easily.
- Everything is driven by a parameter (text) file: you edit or create parameter files to run your experiments.
- No need to be connected to the internet, everything runs locally on the iPhone.
- Additional sample files and example vocabularies available free online.
- Uses free, speaker-independent, open source, connected speech from CMU PocketSphinx project.
- Useful for personal experimentation, classroom and lab instruction, workshop presentations and more.
- Provides appropriate defaults, allowing most designers to use the app off-the-shelf with no modifications to either the vocabulary or the error-repair behaviors.
TaskSpeech apps aim to demystify ASR, emphasizing its strengths and weaknesses as a technology that must be engineered to work properly. The goal is to veer users' thinking away from the notion of artificial speech as magic — or as hopelessly opaque — replacing it with a realistic and practical approach to VUI design. TS+ YesNo is targeted for specific end users and use-cases. Equipped with the app:
- Students can learn about introductory ASR, including grammars and language models, end-pointing, error-types and -recovery, self-teaching turnarounds, and other mechanics.
- Designers can choose between strict enforcement or very loose enforcement of TaskSpeech rules and protocol.
- Teachers can create materials for classes and workshops.
- Experimenters and speech researchers can explore ASR inexpensively, at their own pace.
- Designers can master ASR concepts and HCI principles before applying them to any specific application or platform.
- Project managers can build working prototypes of subdialogues for stakeholder buy-in prior to expensive development on their chosen speech platform.