Wordspotting for voice editing and indexing

Automatic Summarization

“…Early systems used a fixed set of keywords and identified spoken instances of these keywords in the speech stream, a task referred to as "wordspotting" [301]. Further development in this area was devoted to dropping the restriction that the keywords must be specified in advance [122].…”

“…Key examples of work conducted in this era are [230,231] from MIT Lincoln Labs and [301] from Xerox PARC. Modern large-vocabulary continuous speech recognition (LVCSR) had not yet emerged onto the scene, and systems addressed the task of filtering voice messages by using wordspotting techniques, which recognized a small set of words within the speech stream.…”

“…These approaches are aptly referred to as "voice grep" by [241]. Such systems were proposed also in order to handle voice input [301]. As previously mentioned, the application of wordspotting systems to tasks related to searching speech focused on message classification, e.g., [230].…”

Unconstrained keyword spotting using phone lattices with application to spoken document retrieval

“…The particular application of the keyword spotting work described in this paper is to locate instances of desired words in a large collection of voice messages, to identify particular messages matching a user's request (Wilcox & Bush, 1991;Glavitsch & Schäuble, 1992;. In its simplest form, this can serve as an "audio grep" that returns a list of messages containing a particular keyword.…”

Keyword detection in conversational speech utterances using hidden Markov model based continuous speech recognition

“…Some of these systems are based on a two pass procedure that is functionally similar to that used by Bridle (1973). Dynamic programming is performed to compute the likelihood of every portion of a continuous input utterance with respect to each hidden Markov word model (Rohlicek, Russel, Roucos & Gish, 1989;Wilcox & Bush, 1991). In these systems, the likelihoods for an independent background speech model is used as part of a likelihood ratio scoring procedure in a decision rule that is applied as a second stage to the word spotter.…”