Conventional LPC Research Articles

Robust F0 Estimation Based on Complex LPC Analysis for IRS Filtered Noisy Speech

This paper proposes a novel robust fundamental frequency (F0) estimation algorithm based on complex-valued speech analysis for an analytic speech signal. Since analytic signal provides spectra only over positive frequencies, spectra can be accurately estimated in low frequencies. Consequently, it is considered that F0 estimation using the residual signal extracted by complex-valued speech analysis can perform better for F0 estimation than that for the residual signal extracted by conventional real-valued LPC analysis. In this paper, the autocorrelation function weighted by AMDF is adopted for the F0 estimation criterion and four signals; speech signal, analytic speech signal, LPC residual and complex LPC residual, are evaluated for the F0 estimation. Speech signals used in the experiments were an IRS filtered speech corrupted by adding white Gaussian noise or Pink noise whose noise levels are 10, 5, 0, -5[dB]. The experimental results demonstrate that the proposed algorithm based on complex LPC residual can perform better than other methods in noisy environment.

Speech recognition in a noisy car environment based on LP of the one-sided autocorrelation sequence and robust similarity measuring techniques

The performance of the existing speech recognition systems degrades rapidly in the presence of background noise. A novel representation of the speech signal, which is based on Linear Prediction of the One-Sided Autocorrelation sequence (OSALPC), has shown to be attractive for noisy speech recognition because of both its high recognition performance with respect to the conventional LPC in severe conditions of additive white noise and its computational simplicity. The aim of this work is twofold: (1) to show that OSALPC also achieves a good performance in a case of real noisy speech (in a car environment), and (2) to explore its combination with several robust similarity measuring techniques, showing that its performance improves by using cepstral liftering, dynamic features and multilabeling.

Robust signal selection for linear prediction analysis of voiced speech

This paper investigates a weighted LPC analysis of voiced speech. In view of the speech production model, the weighting function is either chosen to be the short-time energy function of the preemphasized speech sample sequence with certain delays or is obtained by thresholding the short-time energy function. In this method, speech samples are selectively weighted based on how well they match the speech production model. Therefore, the estimates of the LPC coefficients obtained by this novel LPC analysis are more accurate than those obtained from the conventional LPC analysis. They are also less sensitive to the values of the fundamental frequency than conventional LPC.

A Japanese text‐to‐speech conversion system using pitch‐controlled residual wave excitation

A Japanese text‐to‐speech conversion system was developed that could generate fluent speech by concatenating CV and VC units. A new synthesis method was proposed, where residual waves were used as excitation signals for an LPC synthesis filter in all portions of each unit. LPC filter coefficients were calculated by approximating spectral envelopes extracted by the improved cepstral analysis method, which was less affected by pitch frequency than the conventional LPC method. Therefore, the synthetic speech generated by the proposed method had extremely high quality, even when pitch frequencies of the residual waves were widely changed. Moreover, in order to realize natural rhythms, phoneme duration was determined by new rules. The rules were created by taking into account the various contexts (such as preceding and following phonemes and the position of a phoneme in a phrase), and by statistically analyzing a large speech database. A comprehension test using 1000 phonetic balanced words showed that the proposed system achieved 97.4% of high accuracy rate.

Temporal energy contour in isolated word recognition

An analysis of recognition errors in a speaker‐independent digit recognition experiment has demonstrated that the temporal energy contour of the teat tokens is an important feature for recognition and perception. This finding supports the conclusions of other recent studies [Rabiner et al., Proc. ICASSP 17.1.1 (1984)], [Rabiner et al., J. Acoust. Soc. Am. Suppl. 1 75, S93 (1984)]. A conventional LPC analysis with a likelihood ratio distance measure and dynamic time warping was used. This recognizer ignores the temporal energy contour. To help us understand the nature of the recognition errors, we played the test tokens through an LPG synthesizer after time alignment with the correct and with the successful (but incorrect) reference templates. Examples on tape demonstrate that the percept of either the correct or of the successful reference word can be evoked by replacing the input energy contour with the reference energy contour.

An enhanced LPC vocoder with no voiced/Unvoiced switch

A modified LPC system (LPC plus) which requires no voiced/unvoiced switch at the synthesizer is presented in this paper. The excitation functions of the synthesizer filter are modeled to be the sum of the conventional pulse and noise sources. The mixture ratio is estimated from the LPC residual error signal, and this parameter controls the amplitudes of the pulse and noise sources. Since the V/UV switch is eliminated, this system produces robust speech in highly noisy environments, while a conventional LPC system produces degraded speech due to voicing errors. In addition, this technique has been applied to the speech of two simultaneous speakers and produces synthesized speech which is more natural and intelligible than that produced by conventional LPC.

An improved LPC algorithm for voiced-speech synthesis

Current LPC synthesizers have a problem when the driving function is periodic; an output synthesized from LPC analysis has spectral properties which are not identical to those of the analyzed input. These distortions prevent a vocoder from being truly transparent for many voiced sounds. An iterative method, which modifies the conventional LPC coefficients such that the autocorrelation function of the output becomes identical to that of the input, is presented. A methodology for keeping the increased computational cost to a minimum is also to be discussed.

On the use of hidden Markov models for speaker‐independent recognition of isolated words from a medium size vocabulary

Recent work at Bell Laboratories has shown how the theories of LPC Vector Quantization (VQ) and hidden Markov modeling (HMM) can be applied to the recognition of isolated word vocabularies. Our first experiments with HMM based recognizers were restricted to a vocabulary of the ten digits. For this simple vocabulary we found that a high performance recognizer (word accuracy on the order of 97%) could be implemented, and that the performance was, for the most part, insensitive to parameters of both the Markov model and the vector quantizer. In this talk we extend our investigations to the recognition of isolated words from a medium size vocabulary, (129 words), as used in the Bell Laboratories airline reservation and information system. For this moderately complex vocabulary we have found that recognition accuracy is indeed a function of the HMM parameter (i.e., the number of states and the number of symbols in the vector quantizer). We have also found that a vector quantizer which uses energy information gives better performance than a conventional LPC shape vector quantizer of the same size (i.e., number of codebook entries).

Frequency weighted LPC spectral matching measures

AbstractSpectral matching is a powerful method of calculating the distance between two frames in speech data which is used in speech analysis, speech recognition and the evaluation of speech quality. Since the introduction of LPC [linear prediction coefficient] analysis, several LPC spectral matching measures have been proposed. It is widely known that human hearing has greater sensitivity in the low‐frequency region than in the high‐frequency region. However, it is impossible to assign frequency weighting using conventional LPC spectral matching measures. This paper proposes two types of frequency weighted LPC spectral matching. The first method is applicable only to the peak weighted measure, while the second is applicable to a more general LPC spectral matching measure. The theoretical relation between these two types of frequency weighting methods is discussed and vowel identification experiments are carried out to examine the effectiveness of these methods. It is seen that the first method is effective in assigning the same frequency weighting to all frames and the second method is effective in assigning different frequency weightings to frames with different phonemic reference patterns.

LPC peak weighted spectral matching measures

AbstractThis report proposed an LPC peak weighted spectral matching measure, a voice recognition method with weighted global spectral peaks (formants), and demonstrates its usefulness by comparing it with the conventional LPC matching measure. The “peak weighted measure” is defined by the integral of the product Kw, where K is a function representing spectral differences between the standard pattern and input voice and w is a weighted function representing peaks. It is shown that the integral value can be computed simply by LPC analysis parameters. Analysis of the weighted spectral difference Kw shows some variation in effectiveness of peak weighting.Experiments on vowel discrimination to evaluate the measure conclude that the discrimination error rate of 11.7% in the conventional LPC matching measure can be reduced to 9.3% by using the peak weighted measure and, especially, it is remarkably improved for the vowels /a/, /u/, /o/. Further, the variance of the discrimination error rates by speakers is smaller than that of the conventional LPC matching measure.

Analysis of the performance of a tandem connection of a 2.4-kB/sec LPC vocoder to a 16-kB/sec CVSD coder

The performance of a tandem connection of narrow-band and wide-band speech communication systems is evaluated. Specifically, the narrow-band system consists of a conventional LPC vocoder operating at a bit rate of 2.4 Kb/sec and the wideband system consists of a CVSD (continuously variable slope delta modulator) operating at a bit rate of 16 Kb/sec. This paper describes the properties of the narrowband to wideband link. The SNR (signal-to-quantizing noise ratio) of the CVSD coder is analyzed over a 50-dB variation of the input signal levels and for a variety of source excitations for the LPC synthesizer. It is shown that SNR improvements in the CVSD coder of 2–2.5 dB are possible in the slope overload region of the coder by modifying the source excitation of the LPC synthesizer and by preprocessing the input signal to the coder with an allpass filter. Both methods aid in reducing the peak factor (peak to rms level) of the input speech to the coder. Subjectively, however, only slight improvements in quality, if any, were observed with these modifications. The weakest link in terms of quality in the tandem connection appears to be that of the CVSD coder.