Claude Barras

Transcriber: Development and use of a tool for assisting speech corpora production

Abstract We present “Transcriber”, a tool for assisting in the creation of speech corpora, and describe some aspects of its development and use. Transcriber was designed for the manual segmentation and transcription of long duration broadcast news recordings, including annotation of speech turns, topics and acoustic conditions. It is highly portable, relying on the scripting language Tcl/Tk with extensions such as Snack for advanced audio functions and tcLex for lexical analysis, and has been tested on various Unix systems and Windows. The data format follows the XML standard with Unicode support for multilingual transcriptions. Distributed as free software in order to encourage the production of corpora, ease their sharing, increase user feedback and motivate software contributions, Transcriber has been in use for over a year in several countries. As a result of this collective experience, new requirements arose to support additional data formats, video control, and a better management of conversational speech. Using the annotation graphs framework recently formalized, adaptation of the tool towards new tasks and support of different data formats will become easier.

Multistage speaker diarization of broadcast news

This paper describes recent advances in speaker diarization with a multistage segmentation and clustering system, which incorporates a speaker identification step. This system builds upon the baseline audio partitioner used in the LIMSI broadcast news transcription system. The baseline partitioner provides a high cluster purity, but has a tendency to split data from speakers with a large quantity of data into several segment clusters. Several improvements to the baseline system have been made. First, the iterative Gaussian mixture model (GMM) clustering has been replaced by a Bayesian information criterion (BIC) agglomerative clustering. Second, an additional clustering stage has been added, using a GMM-based speaker identification method. Finally, a post-processing stage refines the segment boundaries using the output of a transcription system. On the National Institute of Standards and Technology (NIST) RT-04F and ESTER evaluation data, the multistage system reduces the speaker error by over 70% relative to the baseline system, and gives between 40% and 50% reduction relative to a single-stage BIC clustering system

Feature and score normalization for speaker verification of cellular data

This paper presents some experiments with feature and score normalization for text-independent speaker verification of cellular data. The speaker verification system is based on cepstral features and Gaussian mixture models with 1024 components. The following methods, which have been proposed for feature and score normalization, are reviewed and evaluated on cellular data: cepstral mean subtraction (CMS), variance normalization, feature warping, T-norm, Z-norm and the cohort method. We found that the combination of feature warping and T-norm gives the best results on the NIST 2002 test data (for the one-speaker detection task). Compared to a baseline system using both CMS and variance normalization and achieving a 0.410 minimal decision cost function (DCF), feature warping and T-norm respectively bring 8% and 12% relative reductions, whereas the combination of both techniques yields a 22% relative reduction, reaching a DCF of 0.320. This result approaches the state-of-the-art performance level obtained for speaker verification with land-line telephone speech.

Constrained MLLR for Speaker Recognition

One particularly difficult challenge for cross-channel MLLR (CMLLR) are two widely-used techniques for speaker introduced in the 2005 and 2006 NIST Speaker Recognition Evaluations, where training uses telephone speech and verification uses speech from multiple auxiliary comparable to that obtained with cepstral features. This paper describes a new feature extraction technique for speaker recognition based on CMLLR speaker adaptation which session effects through latent factor analysis (LFA) and through support vector machines (SVM). Results on the NIST operates directly on the recorded signal with noise well as in combination with two cepstral approaches such as reduction in the performance gap between telephone and auxiliary microphone data.

Speaker diarization: from broadcast news to lectures

This paper presents the LIMSI speaker diarization system for lecture data, in the framework of the Rich Transcription 2006 Spring (RT-06S) meeting recognition evaluation. This system builds upon the baseline diarization system designed for broadcast news data. The baseline system combines agglomerative clustering based on Bayesian information criterion with a second clustering using state-of-the-art speaker identification techniques. In the RT-04F evaluation, the baseline system provided an overall diarization error of 8.5% on broadcast news data. However since it has a high missed speech error rate on lecture data, a different speech activity detection approach based on the log-likelihood ratio between the speech and non-speech models trained on the seminar data was explored. The new speaker diarization system integrating this module provides an overall diarization error of 20.2% on the RT-06S Multiple Distant Microphone (MDM) data.

Comparison of Speaker Adaptation Methods as Feature Extraction for SVM-Based Speaker Recognition

In the last years the speaker recognition field has made extensive use of speaker adaptation techniques. Adaptation allows speaker model parameters to be estimated using less speech data than needed for maximum-likelihood (ML) training. The maximum a posteriori (MAP) and maximum-likelihood linear regression (MLLR) techniques have typically been used for adaptation. Recently, MAP and MLLR adaptation have been incorporated in the feature extraction stage of support vector machine (SVM)-based speaker recognition systems. Two approaches to feature extraction use a SVM to classify either the MAP-adapted Gaussian mean vector parameters (GSV-SVM) or the MLLR transform coefficients (MLLR-SVM). In this paper, we provide an experimental analysis of the GSV-SVM and MLLR-SVM approaches. We largely focus on the latter by exploring constrained and unconstrained transforms and different choices of the acoustic model. A channel-compensated front-end is used to prevent the MLLR transforms to adapt to channel components in the speech data. Additional acoustic models were trained using speaker adaptive training (SAT) to better estimate the speaker MLLR transforms. We provide results on the NIST 2005 and 2006 Speaker Recognition Evaluation (SRE) data and fusion results on the SRE 2006 data. The results show that using the compensated front-end, SAT models and multiple regression classes bring major performance improvements.

Multi-stage Speaker Diarization for Conference and Lecture Meetings

The LIMSI RT-07S speaker diarization system for the conference and lecture meetings is presented in this paper. This system builds upon the RT-06S diarization system designed for lecture data. The baseline system combines agglomerative clustering based on Bayesian information criterion (BIC) with a second clustering using state-of-the-art speaker identification (SID) techniques. Since the baseline system provides a high speech activity detection (SAD) error around of 10% on lecture data, some different acoustic representations with various normalization techniques are investigated within the framework of log-likelihood ratio (LLR) based speech activity detector. UBMs trained on the different types of acoustic features are also examined in the SID clustering stage. All SAD acoustic models and UBMs are trained with the forced alignment segmentations of the conference data. The diarization system integrating the new SAD models and UBM gives comparable results on both the RT-07S conference and lecture evaluation data for the multiple distant microphone (MDM) condition.

The LIMSI 2006 TC-STAR EPPS Transcription Systems

This paper describes the speech recognizers developed to transcribe European Parliament Plenary Sessions (EPPS) in English and Spanish in the 2nd TC-STAR Evaluation Campaign. The speech recognizers are state-of-the-art systems using multiple decoding passes with models (lexicon, acoustic models, language models) trained for the different transcription tasks. Compared to the LIMSI TC-STAR 2005 EPPS systems, relative word error rate reductions of about 30% have been achieved on the 2006 development data. The word error rates with the LIMSI systems on the 2006 EPPS evaluation data are 8.2% for English and 7.8% for Spanish. Experiments with cross-site adaptation and system combination are also described.

Automatic transcription of compressed broadcast audio

With increasing volumes of audio and video data broadcast over the Web, it is of interest to assess the performance of state-of-the-art automatic transcription systems on compressed audio data for media indexation applications. In this paper the performance of the LIMSI 10x French broadcast news transcription system is measured on a two-hour audio set for a range of MP3 and RealAudio codecs at various bit rates and the GSM codec used for European cellular phone communications. The word error rates are compared with those obtained on high quality PCM recordings prior to compression. For a 6.5 kbps. audio bit rate (the most commonly used on the Web), word error rates under 40% can be achieved, which makes automatic media monitoring systems over the Web a realistic task.

Impact of overlapping speech detection on speaker diarization for broadcast news and debates

The overlapping speech detection systems developped by Orange and LIMSI for the ETAPE evaluation campaign on French broadcast news and debates are described. Using either cepstral features or a multi-pitch analysis, a F1-measure for overlapping speech detection up to 59.2% is reported on the TV data of the ETAPE evaluation set, where 6.7% of the speech was measured as overlapping, ranging from 1.2% in the news to 10.4% in the debates. Overlapping speech segments were excluded during the speaker diarization stage, and these segments were further labelled with the two nearest speaker labels, taking into account the temporal distance. We describe the effects of this strategy for various overlapping speech systems and we show that it improves the diarization error rate in all situations and up to 26.1% relative in our best configuration.