Spoken language recognition (SLR) refers to the automatic process used to determine the language present in a speech sample. SLR is an important task in its own right, for example, as a tool to analyze or categorize large amounts of multi-lingual data. Further, it is also an essential tool for selecting downstream applications in a work flow, for example, to chose appropriate speech recognition or machine translation models. SLR systems are usually composed of two stages, one where an embedding representing the audio sample is extracted and a second one which computes the final scores for each language. In this work, we approach the SLR task as a detection problem and implement the second stage as a probabilistic linear discriminant analysis (PLDA) model. We show that discriminative training of the PLDA parameters gives large gains with respect to the usual generative training. Further, we propose a novel hierarchical approach where two PLDA models are trained, one to generate scores for clusters of highly-related languages and a second one to generate scores conditional to each cluster. The final language detection scores are computed as a combination of these two sets of scores. The complete model is trained discriminatively to optimize a cross-entropy objective. We show that this hierarchical approach consistently outperforms the non-hierarchical one for detection of highly related languages, in many cases by large margins. We train our systems on a collection of datasets including over 100 languages, and test them both on matched and mismatched conditions, showing that the gains are robust to condition mismatch.

In this work, we assess the impact of vocal effort on discrimination and calibration performance of a state-of-the-art speaker recognition system. We analyze three levels of vocal effort (low, normal, and high) from the SRI-FRTIV corpus. We use a deep neural network (DNN) speaker embeddings system with probabilistic linear discriminant analysis (PLDA) and find that vocal effort variation significantly degrades system performance. We apply both mixture PLDA (mix-PLDA) and trial-based calibration with condition PLDA similarity (TBC-CPLDA) to improve system robustness. Our proposed approaches resulted in 18% and 33% relative improvement in discrimination and calibration performance respectively on the SRI-FRTIV corpus.

The output scores of most of the speaker recognition systems are not directly interpretable as stand-alone values. For this reason, a calibration step is usually performed on the scores to convert them into proper likelihood ratios, which have a clear probabilistic interpretation. The standard calibration approach transforms the system scores using a linear function trained using data selected to closely match the evaluation conditions. This selection, though, is not feasible when the evaluation conditions are unknown. In previous work, we proposed a calibration approach for this scenario called trial-based calibration (TBC). TBC trains a separate calibration model for each test trial using data that is dynamically selected from a candidate training set to match the conditions of the trial. In this work, we extend the TBC method, proposing: 1) a new similarity metric for selecting training data that result in significant gains over the one proposed in the original work; 2) a new option that enables the system to reject a trial when not enough matched data are available for training the calibration model; and 3) the use of regularization to improve the robustness of the calibration models trained for each trial. We test the proposed algorithms on a development set composed of several conditions and on the Federal Bureau of Investigation multi-condition speaker recognition dataset, and we demonstrate that the proposed approach reduces calibration loss to values close to 0 for most of the conditions when matched calibration data are available for selection, and that it can reject most of the trials for which relevant calibration data are unavailable.

This paper introduces the Voices Obscured In Complex Environmental Settings (VOICES) corpus, a freely available dataset under Creative Commons BY 4.0.This dataset will promote speech and signal processing research of speech recorded by far-field microphones in noisy room conditions.Publicly available speech corpora are mostly composed of isolated speech at close-range microphony.A typical approach to better represent realistic scenarios, is to convolve clean speech with noise and simulated room response for model training.Despite these efforts, model performance degrades when tested against uncurated speech in natural conditions.For this corpus, audio was recorded in furnished rooms with background noise played in conjunction with foreground speech selected from the Lib-riSpeech corpus.Multiple sessions were recorded in each room to accommodate for all foreground speech-background noise combinations.Audio was recorded using twelve microphones placed throughout the room, resulting in 120 hours of audio per microphone.This work is a multi-organizational effort led by SRI International and Lab41 with the intent to push forward state-of-the-art distant microphone approaches in signal processing and speech recognition.

Using bottleneck features extracted from a deep neural network (DNN) trained to predict senone posteriors has resulted in new, state-of-the-art technology for language and speaker identification. For language identification, the features' dense phonetic information is believed to enable improved performance by better representing language-dependent phone distributions. For speaker recognition, the role of these features is less clear, given that a bottleneck layer near the DNN output layer is thought to contain limited speaker information. In this article, we analyze the role of bottleneck features in these identification tasks by varying the DNN layer from which they are extracted, under the hypothesis that speaker information is traded for dense phonetic information as the layer moves toward the DNN output layer. Experiments support this hypothesis under certain conditions, and highlight the benefit of using a bottleneck layer close to the DNN output layer when DNN training data is matched to the evaluation conditions, and a layer more central to the DNN otherwise.

The newly collected Speakers in the Wild (SITW) database was central to a text-independent speaker recognition challenge held as part of a special session at Interspeech 2016. The SITW database is composed of audio recordings from 299 speakers collected from open source media, with an average of 8 sessions per speaker. The recordings contain unconstrained or wild acoustic conditions, rarely found in large speaker recognition datasets, and multi-speaker recordings for both speaker enrollment and verification. This article provides details of the SITW speaker recognition challenge and analysis of evaluation results. There were 25 international teams involved in the challenge of which 11 teams participated in an evaluation track. Teams were tasked with applying existing and novel speaker recognition algorithms to the challenges associated with the real world conditions of SITW. We provide an analysis of some of the top performing systems submitted during the evaluation and provide future research directions. Copyright ©2016 ISCA.

We introduce a new dataset for the study of the effect of highly non-stationary noises on language recognition (LR) performance. The dataset is based on the data from the 2009 Language Recognition Evaluation organized by the National Institute of Standards and Technology (NIST). Randomly selected noises are added to these signals to achieve a chosen signal-tonoise ratio and percentage of corruption. We study the effect of these noises on LR performance as a function of these parameters and present some initial methods to mitigate the degradation, focusing on the speech activity detection (SAD) step. These methods include discarding the C0 coefficient from the features used for SAD, using a more stringent threshold on the SAD scores, thresholding the speech likelihoods returned by the model as an additional way of detecting noise, and a final model adaptation step. We show that a system optimized for clean speech is clearly suboptimal on this new dataset since the proposed methods lead to gains of up to 35% on the corrupted data, without knowledge of the test noises and with very little effect on clean data performance. Index Terms: spoken language recognition, non-stationary noise, speech activity detection

This paper proposes adaptive Gaussian backend (AGB), a novel approach to robust language identification (LID). In this approach, a given test sample is compared to language-specific training data in order to dynamically select data for a trialspecific language model. Discriminative AGB additionally weights the training data to maximize discrimination against the test segment. Evaluated on heavily degraded speech data, discriminative AGB provides relative improvements of up to 45% and 38 % in equal error rates (EER) over the widely adopted Gaussian backend (GB) and neural network (NN) approaches to LID, respectively. Discriminative AGB also significantly outperforms those techniques at shorter test durations, while demonstrating robustness to limited training resources and to mismatch between training and testing speech duration. The efficacy of AGB is validated on clean speech data from National Institute of Standards and Technology (NIST) language recognition evaluation (LRE) 2009, on which it was found to provide improvements over the GB and NN approaches. Index Terms: language recognition, adaptive Gaussian backend, support vector machines, noisy speech 1.

As automatic speech processing has matured, research attention has expanded to paralinguistic speech problems that aim to detect beyond-the-words information. This paper focuses on the identification of seven speaker trait categories from the Interspeech Speaker Trait Challenge: likeability, intelligibility, openness, conscientiousness, extraversion, agreeableness, and neuroticism. Our approach combines multiple features including prosodic, cepstral, shifted-delta cepstral, and a reduced set of the OpenSMILE features. Our classification approaches included GMM-UBM, eigenchannel, support vector machines, and distance based classifiers. Optimized feature reduction and logistic regression-based score calibration and fusion led to results that perform competitively against the challenge baseline in all categories.

This study seeks to quantify the effectiveness of a broad range of acoustic features for speaker identification and their impact in feature fusion. Sixteen different acoustic features are evaluated under nine different acoustic, channel and speaking style conditions. Three major types of features are examined: traditional (MFCC, PLP, LPCC, etc.), innovative (PYKFEC, MVDR, etc.) and extensions of these (frequency-constrained LPCC, LFCC). All features were then fused in binary and three-way fusion to determine the complementarity between features and their impact on accuracy. Results were surprising, with the MVDR feature having the highest performance for any single feature, and LPCC based features having the greatest impact on fusion effectiveness. Commonly used features like PLP and MFCC did not achieve the best results in any category. It was further found that removing the perceptually-motivated warping from MFCC, MVDR and PYKFEC improved the performance of these features significantly.

Nearest neighbor searching (NNS) is a common classification method, but its brute-force (BF) implementation is inefficient for dimensions greater than 10. We present Cellular Class Encoding (CCE), shown to be 1.1–1.7 times faster than BF on real-world, 14-dimensional data sets. Moreover, if applied to bounded sets, CCE is a full-search equivalent to BF. Given a query in an indexed cell of a partitioned bounded space, the CCE's efficiency is achieved by only performing NNS on those database elements which could not be eliminated a priori as impossible nearest neighbors of that cell's resident vectors. To ensure CCE is a viable alternative in real-world applications, we use VQ speaker identification as a testbed application and present results.

This study proposes an approach to improving speaker recognition through the process of minute vocal tract length perturbation of training files, coupled with pitch normalization for both train and test data. The notion of perturbation as a method for improving the robustness of training data for supervised classification is taken from the field of optical character recognition, where distorting characters within a certain range has shown strong improvements across disparate conditions. This paper demonstrates that acoustic perturbation, in this case analysis, distortion, and resynthesis of vocal tract length for a given speaker, significantly improves speaker recognition when the resulting files are used to augment or replace the training data. A pitch length normalization technique is also discussed, which is combined with perturbation to improve open-set speaker recognition from an EER of 20% to 6.7%.

Abstract : This project describes the composition and goals of the Multi-session Audio Research Project (MARP) corpus and some initial experimental findings. The MARP corpus is a three year longitudinal collection of 21 sessions and more than 60 participants. This study was undertaken to test the impact of various factors on speaker recognition, such as inter-session variability, intonation, aging, whispering and text dependency. Initial results demonstrate the impact of sentence intonation, whispering, text dependency and cross session tests. These results highlight the sensitivity of speaker recognition to vocal, environmental and phonetic conditions that are commonly encountered but rarely explored or tested.

Session variability in speaker recognition is a well recognized phenomena, but poorly understood largely due to a dearth of robust longitudinal data. The current study uses a large, longterm speaker database to quantify both speaker variability changes within a conversation and the impact of speaker variability changes over the long term (3 years). Results demonstrate that 1) change in accuracy over the course of a conversation is statistically very robust and 2) that the aging effect over three years is statistically negligible. Finally we demonstrate that voice change during the course of a conversation is, in large part, comparable across sessions. Index Terms: session variability, speaker recognition, speaker variability analysis, conversation analysis

Abstract : This paper examines the impact of lossy speech coding with Speex on GMM-UBM speaker recognition (SR). Audio from 120 speakers was compressed with Speex into twelve data sets, each with a different level of compression quality from 0 (most compressed) to 10 (least), plus uncompressed. Experiments looked at performance under matched and mismatched compression conditions, using models conditioned for the coded environment, and Speex coding applied to improving SR performance on other coders. Results show that Speex is effective for compression of data used in SR and that Speex coding can improve performance on data compressed by the GSM codec.

This paper describes the speech database that has a central role in the Interspeech 2007 special session "Novel techniques for the NATO non-native Air Traffic Communications database." The rationale for recording and distributing this common research object is given, and details about the acquisition and annotation are given, as well as some statistics. Further, a summary is given of potential uses of the database, in terms of evaluation measures and protocols.