Efficient subword lattice retrieval for German spoken term detection

Abstract

We present a lattice-based STD method for German broadcast news data and compare it to a previously proposed fuzzy search. Due to the important out-of-vocabulary (OOV) problem in German, we evaluate suitable subword indexing units for lattice retrieval. Hybrid lattice retrieval of words and subwords is investigated because of the robust nature of words as an indexing unit. We show that by using efficient lattice graph and score pruning techniques, precision of subword retrieval is increased by 8% absolute with only a small loss in recall. Additionally, a speed-up of up to 6 times can be observed

Automatic Summarization

“…In [184], two hybrid approaches for STD that combine syllables and word ASR output are proposed. The first approach combines words with fuzzy search in a 1-best syllable transcript.…”

Stochastic pronunciation modelling and soft match for out-of-vocabulary spoken term detection

“…A typical STD system comprises an ASR subsystem for lattice generation and a STD subsystem for term detection, as illustrated in Figure 1. State-of-the-art STD systems include those reported in [2,3,4,5,6,7]. Fig.…”

Robust spoken term detection using partial search and re-scoring hypothesized detections techniques

“…By doing so, the OOV problem is mostly solved since subword dictionaries have much higher coverage than word-based dictionaries. Some typical subword units are phones [56][57][58][59][60][61], syllables [62][63][64], morpheme [65].…”

“…The first strategy, referred to as query expansion [66,67], transforms the keyword into other word/subword sequences that might be confused with the actual keyword, then use them additionally to perform retrieval. The second strategy, referred to as fuzzy matching [57,63,68], estimates the distance between the subword sequence of the keyword and a lattice, and produces a matching if the distance is lower than a specific threshold. In these strategies, a confusion model that encodes the insertion/deletion/substitution costs between subword units is required.…”

“…A wide range of subword units have been proposed, including linguistical units such as phones [56-61, 92, 98, 116], syllables [62][63][64] and morpheme [65], as well as data-driven units called word-fragment [117], multigrams [94,118], graphone [107] and particles [101].…”