EditorialEditorial for the special issue on spoken content retrieval
Introduction
Increasing computing power, storage space and connection bandwidth make large amounts of multimedia data widely available. In addition to text and metadata-based search, content-based retrieval is emerging as a means to provide easy access to multimedia, especially for data containing speech. However, so far retrieval of spoken content has been restricted mostly to the accompanying text meta-data. As research in speech recognition has matured, there has been a shift from development of core technology to applications such as retrieval of spoken content. Unlocking the spoken content holds great promise for improving such applications. This special issue brings together some of the current approaches to spoken content retrieval with the goal of offering a glimpse into the future of the area.
Over the last decade, significant advances have resulted from a combination of advances in automatic speech recognition (ASR), and information retrieval (IR). Innovative methods for tighter coupling of the component technologies together with novel work in query-by-example have been crucial for robustness to errors in speech retrieval. Moving towards open vocabulary search is necessitated by the increased presence of many heterogenous sources. This requires detection and recovery of out-of-vocabulary (OOV) terms as a first step in the discovery of relations to known terms and topics in information extraction. OOV query search is a critical component of the keyword search systems used in DARPA RATS and IARPA Babel programs, as well as the NIST OpenKWS evaluations. Summary of the previous work in the field can be found in Chelba et al., 2008, Chelba et al., 2011.
Section snippets
Summary of the articles in this special issue
This special issue consists of four articles that explore various aspects of spoken content retrieval.
In Eskevich and Jones (2014), the authors explore retrieval of informal speech from meetings assuming a recall-focused search scenario where a user is potentially looking for all relevant topical segments. Search of meetings is an interesting task for speech retrieval since it incorporates many challenges: (i) the content may be spoken in a wide range of often informal spontaneous styles (ii)
References (6)
- et al.
Exploring speech retrieval from meetings using the AMI corpus
Comput. Speech Lang.
(2014) - et al.
Improved open-vocabulary spoken content retrieval with word and subword lattices using acoustic feature similarity
Comput. Speech Lang.
(2014) - et al.
Language independent search in MediaEval's spoken web search task
Comput. Speech Lang.
(2014)