Declarative markup languages for speech applications such as VoiceXML are becoming more prevalent programming modalities for describing speech applications. Present declarative markup languages for speech applications model the running speech application as a state machine with the program specifying the transitions amongst the states. These languages can be extended to support a marker-semantic to more easily solve several problems that are otherwise not easily solved. In one embodiment, a partially overlapping target window is implemented using a mark semantic. Other uses include measurement of user listening time, detection and avoidance of errors, and better resumption of playback after a false barge in.

A device-independent speech audio system for linking a speech driven application to specific audio input and output devices can include a media framework for transporting digitized speech audio between speech driven applications and a plurality of audio input and output devices. The media framework can include selectable device-dependent parameters which can enable the transportation of the digitized speech to and from the plurality of audio input and output devices. The device-independent speech audio system also can include an audio abstractor configurable to provide specific ones of the selectable device-dependent parameters according to the specific audio input and output devices. Hence, the audio abstractor can provide a device-independent interface to the speech driven application for linking the speech driven application to the specific audio input and output devices.

Techniques are described for employing a set of tags to model phenomena which are smooth and subject to constraints. Tags may be used to model, for example, muscular movement producing speech. In one advantageous application, a set of tags defining prosodic characteristics is developed, and selected tags are placed in appropriate locations of a body of text. Each tag defines a constraint on the prosodic characteristics of speech produced by processing the text. Processing of the body of speech and the tags produces a set of equations which are solved to produce a curve defining prosodic characteristics over the scope of a phrase, and a further set of equations which are solved to produce a curve defining prosodic characteristics of individual words within a phrase. The data defined by the curves is used with the text to produce speech having the prosodic characteristics defined by the tags. A set of tags may be produced by reading of a training text by a target speaker to produce a training corpus reflecting the prosodic characteristics of the target speaker, and then analyzing the training corpus to generate tags modeling the prosodic characteristics of the training corpus.

Techniques for generating, distributing, and using speech recognition models are described. A shared speech processing facility is used to support speech recognition for a wide variety of devices with limited capabilities including business computer systems, personal data assistants, etc., which are coupled to the speech processing facility via a communications channel, e.g., the Internet. Devices with audio capture capability record and transmit to the speech processing facility, via the Internet, digitized speech and receive speech processing services, e.g., speech recognition model generation and/or speech recognition services, in response. The Internet is used to return speech recognition models and/or information identifying recognized words or phrases. Thus, the speech processing facility can be used to provide speech recognition capabilities to devices without such capabilities and/or to augment a device's speech processing capability. Voice dialing, telephone control and/or other services are provided by the speech processing facility in response to speech recognition results.

Declarative markup languages for speech applications such as VoiceXML are becoming more prevalent programming modalities for describing speech applications. Present declarative markup languages for speech applications model the running speech application as a state machine with the program specifying the transitions amongst the states. These languages can be extended to support a marker-semantic to more easily solve several problems that are otherwise not easily solved. In one embodiment, a partially overlapping target window is implemented using a mark semantic. Other uses include measurement of user listening time, detection and avoidance of errors, and better resumption of playback after a false barge in.