Publications
Speaker verification using adapted Gaussian mixture models
January 1, 2000
Journal Article
Published in:
Digit. Signal Process., Vol. 10, No. 1-3, January/April/July, 2000, pp. 19-41. (Fifth Annual NIST Speaker Recognition Workshop, 3-4 June 1999.)
Topic:
R&D area:
Summary
In this paper we describe the major elements of MIT Lincoln Laboratory's Gaussian mixture model (GMM)-based speaker verification system used successfully in several NIST Speaker Recognition Evaluations (SREs). The system is built around the likelihood ratio test for verification, using simple but effective GMMs for likelihood functions, a universal background model (UBM) for alternative speaker representation, and a form of Bayesian adaptation to derive speaker models from the UBM. The development and use of a handset detector and score normalization to greatly improve verification performance is also described and discussed. Finally, representative performance benchmarks and system behavior experiments on NIST SRE corpora are presented.
Summary
In this paper we describe the major elements of MIT Lincoln Laboratory's Gaussian mixture model (GMM)-based speaker verification system used successfully in several NIST Speaker Recognition Evaluations (SREs). The system is built around the likelihood ratio test for verification, using simple but effective GMMs for likelihood functions, a universal background...
READ MORE
Estimation of modulation based on FM-to-AM transduction: two-sinusoid case
November 1, 1999
Journal Article
Published in:
IEEE Trans. Signal Process., Vol. 47, No. 11, November 1999, pp. 3084-3097.
Topic:
R&D area:
Summary
A method is described for estimating the amplitude modulation (AM) and the frequency modulation (FM) of the components of a signal that consists of two AM-FM sinusoids. The approach is based on the transduction of FM to AM that occurs whenever a signal of varying frequency passes through a filter with a nonflat frequency response. The objective is to separate the AM and FM of the sinusoids from the amplitude envelopes of the output of two transduction filters, where the AM and FM are nonlinearly combined in the amplitude envelopes. A current scheme is first refined for AM-FM estimation of a single AM-FM sinusoid by iteratively inverting the AM and FM estimates to reduce error introduced in transduction. The transduction filter pair is designed relying on both a time-and frequency-domain characterization of transduction error. The approach is then extended to the case of two AM-FM sinusoids by essentially reducing the problem to two single-component AM-FM estimation problems. By exploiting the beating in the amplitude envelope of each filter output due to the two-sinusoidal input, a closed-form solution is obtained. This solution is also improved upon by iterative refinement. The AM-FM estimation methods are evaluated through an error analysis and are illustrated for a wide range of AM-FM signals.
Summary
A method is described for estimating the amplitude modulation (AM) and the frequency modulation (FM) of the components of a signal that consists of two AM-FM sinusoids. The approach is based on the transduction of FM to AM that occurs whenever a signal of varying frequency passes through a filter...
READ MORE
Shunting networks for multi-band AM-FM decomposition
October 17, 1999
Conference Paper
Published in:
Proc. IEEE Workshop on Applications of Signal Processing to Audio and Acoustics, 17-20 October 1999.
Topic:
R&D area:
Summary
We describe a transduction-based, neurodynamic approach to estimating the amplitude-modulated (AM) and frequency-modulated (FM) components of a signal. We show that the transduction approach can be realized as a bank of constant-Q bandpass filters followed by envelope detectors and shunting neural networks, and the resulting dynamical system is capable of robust AM-FM estimation. Our model is consistent with recent psychophysical experiments that indicate AM and FM components of acoustic signals may be transformed into a common neural code in the brain stem via FM-to-AM transduction. The shunting network for AM-FM decomposition is followed by a contrast enhancement shunting network that provides a mechanism for robustly selecting auditory filter channels as the FM of an input stimulus sweeps across the multiple filters. The AM-FM output of the shunting networks may provide a robust feature representation and is being considered for applications in signal recognition and multi-component decomposition problems.
Summary
We describe a transduction-based, neurodynamic approach to estimating the amplitude-modulated (AM) and frequency-modulated (FM) components of a signal. We show that the transduction approach can be realized as a bank of constant-Q bandpass filters followed by envelope detectors and shunting neural networks, and the resulting dynamical system is capable of...
READ MORE
Speaker and language recognition using speech codec parameters
September 5, 1999
Conference Paper
Published in:
EUROSPEECH 99, 5-10 September 1999.
Topic:
R&D area:
Summary
In this paper, we investigate the effect of speech coding on speaker and language recognition tasks. Three coders were selected to cover a wide range of quality and bit rates: GSM at 12.2 kb/s, G.729 at 8 kb/s, and G.723.1 at 5.3 kb/s. Our objective is to measure recognition performance from either the synthesized speech or directly from the coder parameters themselves. We show that using speech synthesized from the three codecs, GMM-based speaker verification and phone-based language recognition performance generally degrades with coder bit rate, i.e., from GSM to G.729 to G.723.1, relative to an uncoded baseline. In addition, speaker verification for all codecs shows a performance decrease as the degree of mismatch between training and testing conditions increases, while language recognition exhibited no decrease in performance. We also present initial results in determining the relative importance of codec system components in their direct use for recognition tasks. For the G.729 codec, it is shown that removal of the post- filter in the decoder helps speaker verification performance under the mismatched condition. On the other hand, with use of G.729 LSF-based mel-cepstra, performance decreases under all conditions, indicating the need for a residual contribution to the feature representation.
Summary
In this paper, we investigate the effect of speech coding on speaker and language recognition tasks. Three coders were selected to cover a wide range of quality and bit rates: GSM at 12.2 kb/s, G.729 at 8 kb/s, and G.723.1 at 5.3 kb/s. Our objective is to measure recognition performance...
READ MORE
Modeling of the glottal flow derivative waveform with application to speaker identification
September 1, 1999
Journal Article
Published in:
IEEE Trans. Speech Audio Process., Vol. 7, No. 5, September 1999, pp. 569-586.
Topic:
R&D area:
Summary
An automatic technique for estimating and modeling the glottal flow derivative source waveform from speech, and applying the model parameters to speaker identification, is presented. The estimate of the glottal flow derivative is decomposed into coarse structure, representing the general flow shape, and fine structure, comprising aspiration and other perturbations in the flow, from which model parameters are obtained. The glottal flow derivative is estimated using an inverse filter determined within a time interval of vocal-fold closure that is identified through differences in formant frequency modulation during the open and closed phases of the glottal cycle. This formant motion is predicted by Ananthapadmanabha and Fant to be a result of time-varying and nonlinear source/vocal tract coupling within a glottal cycle. The glottal flow derivative estimate is modeled using the Liljencrants-Fant model to capture its coarse structure, while the fine structure of the flow derivative is represented through energy and perturbation measures. The model parameters are used in a Gaussian mixture model speaker identification (SID) system. Both coarse- and fine-structure glottal features are shown to contain significant speaker-dependent information. For a large TIMIT database subset, averaging over male and female SID scores, the coarse-structure parameters achieve about 60% accuracy, the fine-structure parameters give about 40% accuracy, and their combination yields about 70% correct identification. Finally, in preliminary experiments on the counterpart telephone-degraded NTIMIT database, about a 5% error reduction in SID scores is obtained when source features are combined with traditional mel-cepstral measures.
Summary
An automatic technique for estimating and modeling the glottal flow derivative source waveform from speech, and applying the model parameters to speaker identification, is presented. The estimate of the glottal flow derivative is decomposed into coarse structure, representing the general flow shape, and fine structure, comprising aspiration and other perturbations...
READ MORE
Implications of glottal source for speaker and dialect identification
March 15, 1999
Conference Paper
Published in:
Proc. IEEE Int. Conf. on Acoustics, Speech and Signal Processing, ICASSP, Vol. II, 15-19 March 1999, pp. 813-816.
Topic:
R&D area:
Summary
In this paper we explore the importance of speaker specific information carried in the glottal source. We time align utterances of two speakers speaking the same sentence from the TIMIT database of American English. We then extract the glottal flow derivative from each speaker and interchange them. Through time alignment and this glottal flow transformation, we can make a speaker of a northern dialect sound more like his southern counterpart. We also time align the utterances of two speakers of Spanish dialects speaking the same sentence and then perform the glottal waveform transformation. Through these processes a Peruvian speaker is made to sound more Cuban-like. From these experiments we conclude that significant speaker and dialect specific information, such as noise, breathiness or aspiration, and vocalization, is carried in the glottal signal.
Summary
In this paper we explore the importance of speaker specific information carried in the glottal source. We time align utterances of two speakers speaking the same sentence from the TIMIT database of American English. We then extract the glottal flow derivative from each speaker and interchange them. Through time alignment...
READ MORE
'Perfect reconstruction' time-scaling filterbanks
March 15, 1999
Conference Paper
Published in:
Proc. IEEE Int. Conf. on Acoustics, Speech and Signal Processing, ICASSP, Vol. III, 15-19 March 1999, pp. 945-948.
Topic:
R&D area:
Summary
A filterbank-based method of time-scale modification is analyzed for elemental signals including clicks, sines, and AM-FM sines. It is shown that with the use of some basic properties of linear systems, as well as FM-to-AM filter transduction, "perfect reconstruction" time-scaling filterbanks can be constructed for these elemental signal classes under certain conditions on the filterbank. Conditions for perfect reconstruction time-scaling are shown analytically for the uniform filterbank case, while empirically for the nonuniform constant-Q (gammatone) case. Extension of perfect reconstruction to multi-component signals is shown to require both filterbank and signal-dependent conditions and indicates the need for a more complete theory of "perfect reconstruction" time-scaling filterbanks.
Summary
A filterbank-based method of time-scale modification is analyzed for elemental signals including clicks, sines, and AM-FM sines. It is shown that with the use of some basic properties of linear systems, as well as FM-to-AM filter transduction, "perfect reconstruction" time-scaling filterbanks can be constructed for these elemental signal classes under...
READ MORE
AM-FM separation using shunting neural networks
October 6, 1998
Conference Paper
Published in:
Proc. of the IEEE-SP Int. Symp. on Time-Frequency and Time-Scale Analysis, 6-9 October 1998, pp. 553-556.
Topic:
R&D area:
Summary
We describe an approach to estimating the amplitude-modulated (AM) and frequency-modulated (FM) components of a signal. Any signal can be written as the product of an AM component and an FM component. There have been several approaches to solving the AM-FM estimation problem described in the literature. Popular methods include the use of time-frequency analysis, the Hilbert transform, and the Teager energy operator. We focus on an approach based on FM-to-AM transduction that is motivated by auditory physiology. We show that the transduction approach can be realized as a bank of bandpass filters followed by envelope detectors and shunting neural networks, and the resulting dynamical system is capable of robust AM-FM estimation in noisy environments and over a broad range of filter bandwidths and locations. Our model is consistent with recent psychophysical experiments that indicate AM and FM components of acoustic signals may be transformed into a common neural code in the brain stem via FM-to-AM transduction. Applications of our model include signal recognition and multi-component decomposition.
Summary
We describe an approach to estimating the amplitude-modulated (AM) and frequency-modulated (FM) components of a signal. Any signal can be written as the product of an AM component and an FM component. There have been several approaches to solving the AM-FM estimation problem described in the literature. Popular methods include...
READ MORE
Magnitude-only estimation of handset nonlinearity with application to speaker recognition
May 15, 1998
Conference Paper
Published in:
Proc. of the 1998 IEEE Int. Conf. on Acoustics, Speech and Signal Processing, ICASSP, Vol. II, Speech Processing II; Neural Networks for Signal Processing, 12-15 May 1998, pp. 745-748.
Topic:
R&D area:
Summary
A method is described for estimating telephone handset nonlinearity by matching the spectral magnitude of the distorted signal to the output of a nonlinear channel model, driven by an undistorted reference. The "magnitude-only" representation allows the model to directly match unwanted speech formants that arise over nonlinear channels and that are a potential source of degradation in speaker and speech recognition algorithms. As such, the method is particularly suited to algorithms that use only spectral magnitude information. The distortion model consists of a memoryless polynomial nonlinearity sandwiched between two finite-length linear filters. Minimization of a mean-squared spectral magnitude error, with respect to model parameters, relies on iterative estimation via a gradient descent technique, using a Jacobian in the iterative correction term with gradients calculated by finite-element approximation. Initial work has demonstrated the algorithm's usefulness in speaker recognition over telephone channels by reducing mismatch between high- and low-quality handset conditions.
Summary
A method is described for estimating telephone handset nonlinearity by matching the spectral magnitude of the distorted signal to the output of a nonlinear channel model, driven by an undistorted reference. The "magnitude-only" representation allows the model to directly match unwanted speech formants that arise over nonlinear channels and that...
READ MORE
Audio signal processing based on sinusoidal analysis/synthesis
January 1, 1998
Book Chapter
Published in:
Chapter 9 in Applications of Digital Signal Processing to Audio and Acoustics, 1998, pp. 343-416.
Topic:
R&D area:
Summary
Based on a sinusoidal model, an analysis/synthesis technique is developed that characterizes audio signals, such as speech and music, in terms of the amplitudes, frequencies, and phases of the component sine waves. These parameters are estimated by applying a peak-picking algorithm to the short-time Fourier transform of the input waveform. Rapid changes in the highly resolved spectral components are tracked by using a frequency-matching algorithm and the concept of "birth" and "death" of the underlying sine waves. For a given frequency track, a cubic phase function is applied to the sine-wave generator, whose output is amplitude-modulated and added to sines for other frequency tracks. The resulting synthesized signal preserves the general wave form shape and is nearly perceptually indistinguishable from the original, thus providing the basis for a variety of applications including signal modification, sound splicing, morphing and extrapolation, and estimation of sound characteristics such as vibrato. Although this sine-wave analysis/synthesis is applicable to arbitrary signals, tailoring the system to a specific sound class can improve performance. A source/filter phase model is introduced within the sine-wave representation to improve signal modification, as in time-scale and pitch change and dynamic range compression, by attaining phase coherence where sinewave phase relations are preserved or controlled. A similar method of achieving phase coherence is also applied in revisiting the classical phase vocoder to improve modification of certain signal classes. A second refinement of the sine-wave analysis/synthesis invokes an additive deterministic/stochastic representation of sounds consisting of simultaneous harmonic and aharmonic contributions. A method of frequency tracking is given for the separation of these components, and is used in a number of applications. The sinewave model is also extended to two additively combined signals for the separation of simultaneous talkers or music duets. Finally, the use of sine-wave analysis/synthesis in providing insight for FM synthesis is described, and remaining challenges, such as an improved sine-wave representation of rapid attacks and other transient events, are presented.
Summary
Based on a sinusoidal model, an analysis/synthesis technique is developed that characterizes audio signals, such as speech and music, in terms of the amplitudes, frequencies, and phases of the component sine waves. These parameters are estimated by applying a peak-picking algorithm to the short-time Fourier transform of the input waveform...
READ MORE