Approaches to language identification using Gaussian mixture models and shifted delta cepstral features

By Pedro A. Torres-Carrasquillo, Elliot Singer, M. A. Kohler, Richard J. Greene, Douglas A. Reynolds, J. Deller. Interspeech, 2002

489 citations recorded by the source corpus.

Published results indicate that automatic language identification (LID) systems that rely on multiple-language phone recognition and n-gram language modeling produce the best performance in formal LID evaluations. By contrast, Gaussian mixture model (GMM) systems, which measure acoustic characteristics, are far more efficient computationally but have tended to provide inferior levels of performance. This paper describes two GMM-based approaches to language identification that use shifted delta cepstra (SDC) feature vectors to achieve LID performance comparable to that of the best phone-based systems. The approaches include both acoustic scoring and a recently developed GMM tokenization system that is based on a variation of phonetic recognition and language modeling. System performance is evaluated on both the CallFriend and OGI corpora.

Subfields: Artificial Intelligence, Signal Processing

View Paper: https://www.semanticscholar.org/paper/99382f7ac870469335b2e5e3a1c5ff8d48871712

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Signal Processing

Citations in the corpus

A Divide-and-Conquer Approach for Language Identification Based on Recurrent Neural Networks (2016, methodology): “Then the Shifted Delta Coefficients (SDC) [29] are computed and concatenated to the MFCC vector.”
Frame-by-frame language identification in short utterances using deep neural networks (2015, methodology): “tokenizations [35][3] or, in acoustic approaches, by using shifted-delta-cepstral features [36].”
KALAKA-3: a database for the assessment of spoken language recognition technology on YouTube audios (2015, methodology): “…sets were considered under the iVector approach: (1) the traditional Mel-Filter Cepstral Coefﬁcient / Shifted Delta Cepstrum (MFCC/SDC) features (Torres-Carrasquillo et al. 2002), under a 7-2-3-7 conﬁguration; and (2) the recently introduced log-likelihood ratios of phone posterior…”
The MITLL NIST LRE 2007 language recognition system (2008, methodology): “The framewise cepstra are collected into a 56-dimensional feature vector composed of 7 static coefficients and stacked with the set of shifted delta cepstral (SDC) features produced by applying a 7-1-3-7 SDC scheme [4].”
Neural Network Bottleneck Features for Language Identification (2014, methodology): “We also found, this configuration produced better LID performance than the SDC coefficients [18].”
Spoken Language Recognition: From Fundamentals to Practice (2013, background): “For example, English has a phonetic system that contains 24 consonants and 14 vowels, Mandarin has 21 consonants and ten vowels, and Spanish has 18 consonants and five vowels [147].”