All these terms are defined in the context of over-the-phone speech-recognition applications.
Audio file
Digital sound file that the computer plays to a caller.
Audio icon
A short, sometimes musical, sound.
Back end
A database that provides information to a speech application.
Branding
The process of embodying aspects of a company's corporate identity.
Call flow
The structure of how the system branches based on the callers ' responses.
Call scripts
A list of prompts to be recorded by a voice talent.
Concatenative TTS
A form of text-to-speech that splices pieces of prerecorded human speech.
Confirmation prompt
A type of prompt used to verify that the system correctly understood the caller.
Down sampling
The process of converting high-fidelity recordings into low-fidelity ones, mimicking the sound quality of a typical telephone line.
Failure prompt
A type of prompt used to indicate that the system has failed to help the caller, either due to the lack of caller responses or lack of responses that are "understood" by the recognizer.
Formant TTS
A form of text-to-speech that uses a noise generator and a series of filters to change the noise to make it sound like speech.
Front end
Another name for the speech-recognition application/system.
Full deployment
The last stage of deployment, where the system takes 100% of the intended call volume.
Help prompt
A prompt used to provide more information to a caller, enabling them to better understand how to use the system.
Initial prompt
The first prompt that is played when in a particular state/context.
Morphological analysis
A comparison of similar designs.
Partial deployment
The second stage of deployment, used to examine how the technical performance of the system performs under real-world conditions.
Pilot test
The first stage of deployment, used to examine the user -interface performance.
Quality assurance test
A test performed to ensure that the application performs according to the specification.
Recognition engine/Recognizer
The technology that listens to audio and then compares it to a set of known patterns.
Recoverable error
An error that can be rectified by the user.
Retry prompt
A prompt used to provide the caller assistance when the system has detected human speech but does not understand it.
Speaker-dependent recognizer
A recognizer that only works for a particular person and must be tuned by listening to a person saying known phrases.
Speaker-independent recognizer
A recognizer that performs equally well for a large segment of the population.
Speaker verification
Technology used to match the biometrics of a person's voice to a known voiceprint, primarily used for authentication.
Speech scientist
A person who can, among other things, tune a recognizer to perform better based on a statistical analysis of the recognizer's performance.
State
A context of the dialogue, typically where the system asks a question and listens for an answer. Also called a turn .
Success prompt
A prompt that indicates a user is about to progress to the next state, having completed the state they are in.
Telephony-based speech recognition
Recognition performed by a computer listening to speech over a phone line.
Text-to-speech (TTS)
Technology that converts text stored in a digital form to a spoken utterance, in real time.
Timeout prompt
A prompt used to provide the caller assistance when the system has not detected any sound above a (low) threshold.
Turn
A context of the dialogue, typically where the system asks a question and listens for an answer. Also called a state.
Unrecoverable error
An error that cannot be rectified by the user.
Usability test
A test used to determine the quality of the user interface, typically run with a few people in a controlled environment.
Vision clip
An audio file that sounds like a recorded conversation between a speech-recognition system and a caller, typically used to illustrate the design of a not-yet built system.
