Spoken Language Identification Research Articles

Hybrid Learning Model and Acoustic Approach to Spoken Language Identification Using Machine Learning

Spoken Language Identification (SLId) is the process of identifying the language of an utterance from an anonymous speaker, irrespective of gender, pronunciation and accent. In this paper we present acoustics based learning model for spoken language identification. An acoustic feature representing the short term power spectrum of sound called Mel Frequency Cepstral Coefficients (MFCC) is used as a part of the investigation in this paper. The proposed system uses a combination of Gaussian Mixture Model (GMM) and the Support Vector Machines (SVM) to handle the problem of multi class classification. The model aims at detecting English, Japanese, French, Hindi, and Telugu. A speech corpus was built using speech samples obtained from a plethora of online podcasts and audio books. This corpus comprised of utterances spanning over a uniform duration of 10 seconds. Preliminary results indicate an overall accuracy of 96%. A more comprehensive and rigorous test indicates an overall accuracy of 80%. The acoustic model combined with learning techniques hence proposed proves to be a viable approach for Language Identification

Deep bottleneck features for spoken language identification.

A key problem in spoken language identification (LID) is to design effective representations which are specific to language information. For example, in recent years, representations based on both phonotactic and acoustic features have proven their effectiveness for LID. Although advances in machine learning have led to significant improvements, LID performance is still lacking, especially for short duration speech utterances. With the hypothesis that language information is weak and represented only latently in speech, and is largely dependent on the statistical properties of the speech content, existing representations may be insufficient. Furthermore they may be susceptible to the variations caused by different speakers, specific content of the speech segments, and background noise. To address this, we propose using Deep Bottleneck Features (DBF) for spoken LID, motivated by the success of Deep Neural Networks (DNN) in speech recognition. We show that DBFs can form a low-dimensional compact representation of the original inputs with a powerful descriptive and discriminative capability. To evaluate the effectiveness of this, we design two acoustic models, termed DBF-TV and parallel DBF-TV (PDBF-TV), using a DBF based i-vector representation for each speech utterance. Results on NIST language recognition evaluation 2009 (LRE09) show significant improvements over state-of-the-art systems. By fusing the output of phonotactic and acoustic approaches, we achieve an EER of 1.08%, 1.89% and 7.01% for 30 s, 10 s and 3 s test utterances respectively. Furthermore, various DBF configurations have been extensively evaluated, and an optimal system proposed.

A GMM-BASED HIERARCHICAL AUTOMATIC LANGUAGE IDENTIFICATION SYSTEM FOR INDIAN LANGUAGES

Automatic spoken language identification (LID) is the task of identifying a language from a short utterance of speech by an unknown speaker. This article describes a novel two-level identification system for Indian languages using acoustic features. In the first level, the system identifies the family of the spoken language; the second level aims at identifying the particular language within the corresponding family. The proposed system has been modeled using Gaussian mixture model (GMM) and utilizes the following acoustic features: mel frequency cepstral coefficients (MFCC) and shifted delta cepstrum (SDC). A new database has been created for nine Indian languages. It is shown that a GMM-based LID system using MFCC with delta and acceleration coefficients is performing well, with 80.56% accuracy. The performance accuracy of the GMM-based LID system with SDC is also considerable.

A bayesian logistic regression approach to spoken language identification

This paper presents a novel token-based approach for spoken language identification (LID) using bayesian logistic regression model, which takes into account prior distribution for parameters of logistic regression models in order to avoid overfitting. Speech utterances are first decoded into token sequences, and then we design a hierarchical system which utilizes bayesian logistic regression model to perform LID task on these token sequences. Experiments conducted on the NIST LRE 2007 database show that the proposed approach provides quite competitive performance compared to other state-of-the-art token-based approaches.

Spoken language identification with phonological and lexical models

A language identification (LID) system that uses phonemotactic models in addition to phoneme models to identify languages is described. The proposed LID system is trained and tested using the OGI multilanguage telephone speech database. The continuous density second-order ergodic variable duration hidden Markov phonemic models are trained for each language using a high accuracy phoneme recognition system developed at Bell Laboratories. The phonemotactic models for each language are trained using a text corpora of about ten million words and grapheme to phoneme converters. The language Li of an incoming speech signal x is hypothesized as the one that produced the highest likelihood f(x‖λi)f(λi‖Li) for all the phonemic models λi of a given set with the phonemotactic constraint. Initially, this LID system was trained and evaluated for English/Spanish language identification and the language identification was 83% correct (79% on English and 88% on Spanish). Results for four languages will be presented. The discriminative power of this LID system can be improved by mapping the phoneme lattice onto a syllable or a word sequence using a lexical analyzer and a trigram syllable or word language model The language identification results with and without interfacing the lexical analyzer will be presented.

Spoken language identification by ergodic HMMS and its state sequences

AbstractThis paper describes an automatic text‐ and speaker‐independent language identification method based on hidden Markov models (HMMs) for acoustic features. The hidden Markov modeling is used to represent the phonotactics for each language. Each language has its proper phonotactics.The HMM topology here is a fully structured (ergodic) model wherein any state could transit to all states. Two kinds of HMMs are used: the discrete HMM (DHMM) with the codebook and the continuous density HMM (CHMM). The HMM was trained using both the Baum‐Welch (forward‐backward) algorithm and the Viterbi algorithm. The latter was used for emphasizing the state transition probability.For comparison, experiments also were conducted on the identification using a mixtured Gaussian distribution model with one state. This single‐state Gaussian distribution model gave the same performance as the HMM trained with the Baum‐Welch algorithm. This is because the transition between states which reflects the characteristics of each language does not affect the likelihood scores very much. This problem was addressed by emphasizing the transition probabilities and using the Viterbi algorithm, which resulted in an improvement in the recognition rates. The trigram for optimal state sequence is introduced. Combining it with the HMM produced the best results.