The system was trained in a speaker dependent mode on 28 minutes of speech from each of 8 speakers, and was tested on independent test material for each speaker. The system was tested with three artificial grammars spanning a broad perplexity range. The average performance of the system measured in percent word error was: 1.4% for a pattern grammar of perplexity 9, 7.5% for a word-pair grammar of perplexity 62, and 32.4% for a null grammar of perplexity 1000.>

Statistical language models have been successfully used to improve the performance of continuous speech recognition algorithms. Application of such techniques is difficult when only a small training corpus is available. The authors present an approach for dealing with limited training available from the DARPA resource management domain. An initial training corpus of sentences was abstracted by replacing sentence fragments or phrases with variables. This training corpus of phrase sequences was used to derive parameters of a Markov model. The probability of a word sequence is then decomposed into the probability of possible phrase sequences within each of the phrases. Initial results obtained on 150 utterances from six speakers in the DARPA database indicate that this language modeling technique has potential for improved recognition performance. Furthermore, this approach provides a framework for incorporating linguistic knowledge into statistical language models.>

A probabilistic model called the stochastic segment model is introduced that describes the statistical dependence of all the frames of a speech segment. The model uses a time-warping transformation to map the sequence of observed frames to the appropriate frames of the segment model. The joint density of the observed frames is then given by the joint density of the selected model frames. The automatic training and recognition algorithms are discussed and a few preliminary recognition results are presented.>

In this paper, we describe BYBLOS, the BBN continuous speech recognition system. The system, designed for large vocabulary applications, integrates acoustic, phonetic, lexical, and linguistic knowledge sources to achieve high recognition performance. The basic approach, as described in previous papers [1, 2], makes extensive use of robust context-dependent models of phonetic coarticulation using Hidden Markov Models (HMM). We describe the components of the BYBLOS system, including: signal processing frontend, dictionary, phonetic model training system, word model generator, grammar and decoder. In recognition experiments, we demonstrate consistently high word recognition performance on continuous speech across: speakers, task domains, and grammars of varying complexity. In speaker-dependent mode, where 15 minutes of speech is required for training to a speaker, 98.5% word accuracy has been achieved in continuous speech for a 350-word task, using grammars with perplexity ranging from 30 to 60. With only 15 seconds of training speech we demonstrate performance of 97% using a grammar.

Developing accurate and robust phonetic models for the different speech sounds is a major challenge for high performance continuous speech recognition. In this paper, we introduce a new approach, called the stochastic segment model, for modelling a variable-length phonetic segment X, an L-long sequence of feature vectors. The stochastic segment model consists of 1) time-warping the variable-length segment X into a fixed-length segment Y called a resampled segment, and 2) a joint density function of the parameters of the resampled segment Y, which in this work is assumed Gaussian. In this paper, we describe the stochastic segment model, the recognition algorithm, and the iterative training algorithm for estimating segment models from continuous speech. For speaker-dependent continuous speech recognition, the segment model reduces the word error rate by one third over a hidden Markov phonetic model.

In previous papers, we have described the segment vocoder, which transmits intelligible speech at 300 b/s in speaker-independent mode, i.e., new users need not train the system. As expected for vector quantizers, the storage and computational requirements of the segment vocoder are significantly larger than those of the standard LPC-10 vocoder. In this paper, we describe methods for reducing computational and storage requirements of the segment vocoder and present an algorithm that is implementable in real-time on hardware containing several Digital Signal Processing chips. The DRT score of the simplified algorithm is 78%.

This paper describes the results of our work in designing a system for large-vocabulary word recognition of continuous speech. We generalize the use of context-dependent Hidden Markov Models (HMM) of phonemes to take into account word-dependent coarticulatory effects, Robustness is assured by smoothing the detailed word-dependent models with less detailed but more robust models. We describe training and recognition algorithms for HMMs of phonemes-in-context. On a task with a 334-word vocabulary and no grammar (i.e., a branching factor of 334), in speaker-dependent mode, we show an average reduction in word error rate from 24% using context-independent phoneme models, to 10% when using robust context-dependent phoneme models.

We present a method for word recognition with input speech transduced simultaneously by several sensors in high levels of broadband acoustic background noise. In prior work on single-input multisensor systems, limited success in machine recognition was achieved by linearly combining multiple sensor signals to yield a robust estimate of the speech signal in the presence of noise. In this paper, we demonstrate that improved recognition results are obtained by using all available sensor signals jointly as a vector, which preserves information from all sensors, as input to the decision process. We report on multisensor configurations using close-talking pressure-gradient microphones and accelerometers placed at the throat and nose of the speaker. The recognition error rates obtained by using the joint output vector are 45% lower than the error rates obtained with the best constituent sensor in the multisensor system; single-input multisensor systems, on the other hand, produce error rates that are about equal to the error rates obtained with the best constituent sensor.

In this paper, we report the results of isolated-word speech recognition tests performed with the Verbex 4000 recognizer on speech data collected 1) in 95 dB and 115 dB SPL broad-band acoustic noise typical in a fighter aircraft cockpit and 2) using several single sensors and two-sensor configurations that we reported in ICASSP-85. The two-sensor systems and the gradient microphones we tested produced about the same recognition performance in 95 dB noise. In 115 dB noise, the recognizer failed to train for all but the accelerometer, which produced nearly constant performance in both noise levels. Also, we demonstrate the feasibility of a new speech recognition methodology that uses a parallel system of multiple input signals transduced simultaneously using different sensors. In selected phonetic discrimination tests involving a feature-based approach, a parallel-input multisensor system reduced the discrimination errors to between one-half and one-twelfth of the number produced by a gradient microphone alone.

In this paper, we examine several methods for text-independent speaker identification of telephone speech with limited duration data, The issue addressed is the assessment of channel characteristics, especially linear aspects, and methods for improving speaker identification performance when the speaker to be identified is on a different telephone channel than that data used for training. We show experimental evidence illustrating the cross-channel problem and also show that the direct approach, of using simple channel-invariant features, can discard much speaker dependent information. The methods we have found to be most effective rely on the training process to incorporate channel variability.

We propose a new method of synthesis to be used for the segment vocoder, which transmits intelligible speech at rates below 300 b/s. The earlier segment vocoder applies LPC analysis to input speech, divides it into segments of variable duration, matches each segment with the nearest template from a codebook, concatenates at the receiver the set of nearest templates, and finally synthesizes the resultant sequence of speech frames using LPC synthesis. The quality of such a segment vocoder cannot exceed that of a standard unquantized LPC vocoder, which sounds buzzy due to the pulse/noise excitation used. Alternatively, by beginning with the waveforms (not the spectral representation) corresponding to the set of nearest templates, we can independently modify the pitch, energy, and duration of each template to match those of the input segment. These modified segments are then concatenated to produce the output waveform. We present here methods for high-quality modification of the pitch and duration of a segment of a speech waveform and show how these methods can be applied to improve the quality of the segment vocoder's output speech.

We present a new and simple method for speech rate modification that yields high quality rate-modified speech. Earlier algorithms either required a significant amount of computation for good quality output speech or resulted in poor quality rate-modified speech. The algorithm we describe allows arbitrary linear or nonlinear scaling of the time axis. The algorithm operates in the time domain using a modified overlap-and-add (OLA) procedure on the waveform. It requires moderate computation and could be easily implemented in real time on currently available hardware. The algorithm works equally well on single voice speech, multiple-voice speech, and speech in noise. In this paper, we discuss an earlier algorithm for time-scale modification (TSM), and present both objective and informal subjective results for the new and previous TSM methods.

This paper describes the results of our work in designing a system for phonetic recognition of unrestricted continuous speech. We describe several algorithms used to recognize phonemes using context-dependent Hidden Markov Models of the phonemes. We present results for several variations of the parameters of the algorithms. In addition, we propose a technique that makes it possible to integrate traditional acoustic-phonetic features into a hidden Markov process. The categorical decisions usually associated with heuristic acoustic-phonetic algorithms are replaced by automated training techniques and global search strategies. The combination of general spectral information and specific acoustic-phonetic features is shown to result in more accurate phonetic recognition than either representation by itself.

Quantization, the process of approximating continuous-amplitude signals by digital (discrete-amplitude) signals, is an important aspect of data compression or coding, the field concerned with the reduction of the number of bits necessary to transmit or store analog data, subject to a distortion or fidelity criterion. The independent quantization of each signal value or parameter is termed scalar quantization, while the joint quantization of a block of parameters is termed block or vector quantization. This tutorial review presents the basic concepts employed in vector quantization and gives a realistic assessment of its benefits and costs when compared to scalar quantization. Vector quantization is presented as a process of redundancy removal that makes effective use of four interrelated properties of vector parameters: linear dependency (correlation), nonlinear dependency, shape of the probability density function (pdf), and vector dimensionality itself. In contrast, scalar quantization can utilize effectively only linear dependency and pdf shape. The basic concepts are illustrated by means of simple examples and the theoretical limits of vector quantizer performance are reviewed, based on results from rate-distortion theory. Practical issues relating to quantizer design, implementation, and performance in actual applications are explored. While many of the methods presented are quite general and can be used for the coding of arbitrary signals, this paper focuses primarily on the coding of speech signals and parameters.

In this paper, we examine several methods for text-independent speaker identification and discuss the problem of achieving robust performance. In a previous paper [1], we have described a radio-channel database that contains highly variable speech data of poor quality. Here, we describe several experiments with the radio-channel database leading to the development of robust features and methods. These experiments show that robustness to one degradation may not be sufficient to improve speaker identification accuracy. A robust feature set and modeling and classification method should mitigate the effects of many of the degradations in the data.

We use speaker normalization for vocoding the speech of a new input speaker by using a speaker dependent segment vocoder operating at a very-low bit rate, below 300 b/s. The normalization consists of a spectral transformation, applied on the spectral parameter vector of the reference speaker, which should improve the match between the reference and input speakers. The optimal spectral transformation is determined by an iterative algorithm that is guaranteed to converge to a local optimum, i.e., the quantization error of the segment vocoder is minimized by the normalization algorithm. We demonstrate the general algorithm by deriving a linear least squares solution for the spectral transformation. We present some results on several male and female speakers.

This paper discusses the use of the Hidden Markov Model (HMM) in phonetic recognition. In particular, we present improvements that deal with the problems of modeling the effect of phonetic context and the problem of robust pdf estimation. The effect of phonetic context is taken into account by conditioning the probability density functions (pdfs) of the acoustic parameters on the adjacent phonemes, only to the extent that there are sufficient tokens of the phoneme in that context. This partial conditioning is achieved by combining the conditioned and unconditioned pdfs models with weights that depend on the confidence in each pdf estimate. This combination is shown to result in better performance than either model by itself. We also show that it is possible to obtain the computational advantages of using discrete probability densities without the usual requirement for large amounts of training data.

In this paper we investigate several methods for reducing the bit rate of a segment vocoder [1] by 35% to 150 b/s. In the original vocoder we used a random sample of vectors as a set of templates for vector quantization. We demonstrate in this paper that this random quantizer is near-optimal by comparing it with quantizers that use clustering algorithms for quantizing speech segments. The reduction of the bit rate of the segment vocoder was achieved primarily by using a segment network, i.e., not all segment templates are allowed to follow a given segment template. The spectral continuity of speech is used to determine the subset of templates, that can be used to quantize an input segment. To achieve the low rate of 150 b/s, we also reduced the bit rate for coding pitch, gain, and segment duration. Finally, we present the bit allocation used for transmitting speech at 150 b/s as a single speaker segment vocoder.

In this paper we discuss several algorithms that can be used to reduce the transmission rate for LPC vocoded speech to around 300 to 400 b/s, with only a modest degradation in speech quality relative to that of fixed-rate 2400 b/s LPC vocoders. We limit the discussion to vocoders that transmit information for single frames (as opposed to whole segments of speech). We start with vector quantization, which reduces the bit rate to around 800 b/s accompanied by a significant but tolerable loss in quality relative to a typical fixed-rate 2400 b/s vocoder. Then we reduce the frame rate using one of two techniques: Fixed-Rate Transmission with Variable Interpolation, or Optimal Variable-Frame-Rate Transmission. We also reduce the data rate necessary for the source parameters (pitch, voicing, gain) from 400 b/s to about 100 b/s by taking advantage of their statistical dependence on the spectrum and some perceptual factors. The final result at 300 b/s has a quality comparable to that of the fixed-rate 800 b/s vector quantization vocoder. At 400 b/s, the quality is, in many respects, better than that of the 800 b/s vocoder and comparable to the 2400 b/s LPC vocoder.

In this paper, we present the preliminary performance of four methods for text-independent speaker identification using speech transmitted over radio channels. In a previous paper [1], we showed that for both laboratory-quality and simulated noisy-channel data in a single-session paradigm, new probabilistic classifiers yielded performance superior to that of a minimum distance classifier. We have recently compiled a speech database consisting of speech transmissions over a radio-channel. The lower quality and higher variability of this database differ markedly from the laboratory-quality databases often used in speech processing research. We present preliminary results with the same four methods of text-independent speaker identification using the radio-channel database with several experimental paradigms including multi-session paradigms. These results show that the probabilistic methods perform significantly better than a minimum-distance classifier for the multi-session paradigm.

We present a method that reduces the bit rate of a low rate LPC vocoder by modelling the sequence of quantized spectra by a Markov chain. To minimize the bit rate, one would want to use a high-order chain. Unfortunately, a high-order chain would require an inordinate amount of data for training. We describe in this paper the use of a variable order Markov chain that maximizes the effective use of a given amount of speech data. To reduce the number of states of a high-order chain, we define an equivalence relation on the states, i.e., "similar" states are grouped together in an equivalence class and a single conditional distribution is associated with the equivalence class. We introduce two equivalence relations. In the first, called variable order Markov chain, the equivalence classes represent the most probable states of any order. In the second method, called variable resolution, the equivalence class is obtained by decreasing the quantization accuracy in representing a spectrum that belongs to a more remote past. For an LPC vocoder with 64 possible spectra (using 6-bit vector quantization), the second method is superior to the first and decreases the entropy from 6 bits to 4 bits per spectrum with 256 equivalence classes.

We introduce a new method for very-low-rate vocoding that the input speech as a sequence of variable-length segments. A segment is a by a spectrum of frames, where each frame is represented by a spectrum, pitch and gain. We use an automatic segmentation algorithm to obtain segments with an average duration comparable to that of a phoneme. A segment is quantized as a single block. The distance measure used for quantization incooporates the appropriate time alignment of two segments. We employ a computationally efficient metric that does not use the usual dynamic programming time warping. Two basic vocoders using the above approach of block quantization have been used to transmit intelligible speech at 200 b/s.

Most text-independent speaker identification methods to date depend on the use of some distance metric for classification. In this paper we develop the use of probability density function (pdf) estimation for text-independent speaker identification. We compare the performance of two parametric and one non-parametric pdf estimation methods to one distance classification method that uses the Mahalanobis distance. Under all conditions tested, the pdf estimation methods performed substantially better than the Mahalanobis distance method. The best method is a non-parametric pdf estimation method.

Using ideas from one-dimensional maximum entropy spectral estimation a two-dimensional spectral estimator is derived by extrapolating the two-dimensional sampled autocorrelation (or covariance) function. The method used maximizes the entropy of a set of random variables. The extrapolation (or prediction) process under this maximum entropy condition is shown to correspond to the most random extension or equivalently to the maximization of the mean-square prediction error when the optimum predictor is used. The two-dimensional extrapolation must he terminated by the investigator. The Fourier transform of the extrapolated autocorrelation function is the two-dimensional spectral estimator. Using this method one can apply windowing prior to calculating the spectral estimate. A specific algorithm for estimating the two-dimensional spectrum is presented, and its computational complexity is estimated. The algorithm has been programmed and computer examples are presented.

Using the ideas from one-dimensional (1-D) maximum entropy spectral estimation, we derive a 2-D spectral estimator by actually extrapolating the 2-D sampled autocorrelation function. The method used here is to maximize the entropy of a set of random variables. The extrapolation (prediction) process under the maximum entropy condition is shown to correspond to the most random extension or to the maximization of the mean square prediction error conditioned on using the optimum predictor. The 2-D extrapolation must be terminated by the investigator. The Fourier transform of the extrapolated autocorrelation function is our 2-D spectral estimator. A specific algorithm for estimating the 2-D spectrum is presented. The algorithm has been programmed and computer examples are presented.