Automatic word recognition in cars

Mokbel, Chafic; Chollet, G.

doi:10.1109/89.466660

Cited by 45 publications

(29 citation statements)

References 42 publications

(21 reference statements)

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…In [32], front-end constrained iterative enhancement was shown to improve ASR in noise. However, in [58], back-end noise modeling with sufficient noise data was shown both analytically and experimentally to be preferable for ASR compared to noise suppression techniques.…”

Section: Recognizer With Codebook Of Noisy Modelsmentioning

confidence: 99%

“…This requirement is difficult to meet in real world scenarios where noise level and type may vary continuously, introducing mismatch between the acoustic models typically trained with either clean or partial noise data. Numerous techniques to reduce mismatch have been proposed in two domains: 1) transformation of noisy speech towards clean, such as noise suppression [32], [54], [58]; 2) noise modeling in the ASR back-end, such as updating acoustic models for noise characteristics [58], using two-dimensional HMMs to decompose speech and noise components [59], and parallel model combination (PMC) [60]. In [32], front-end constrained iterative enhancement was shown to improve ASR in noise.…”

Section: Recognizer With Codebook Of Noisy Modelsmentioning

confidence: 99%

See 1 more Smart Citation

Unsupervised Equalization of Lombard Effect for Speech Recognition in Noisy Adverse Environments

Bořil

Hansen

2010

IEEE Trans. Audio Speech Lang. Process.

View full text Add to dashboard Cite

Abstract-In the presence of environmental noise, speakers tend to adjust their speech production in an effort to preserve intelligible communication. The noise-induced speech adjustments, called Lombard effect (LE), are known to severely impact the accuracy of automatic speech recognition (ASR) systems. The reduced performance results from the mismatch between the ASR acoustic models trained typically on noise-clean neutral (modal) speech and the actual parameters of noisy LE speech. In this study, novel unsupervised frequency domain and cepstral domain equalizations that increase ASR resistance to LE are proposed and incorporated in a recognition scheme employing a codebook of noisy acoustic models. In the frequency domain, short-time speech spectra are transformed towards neutral ASR acoustic models in a maximum-likelihood fashion. Simultaneously, dynamics of cepstral samples are determined from the quantile estimates and normalized to a constant range. A codebook decoding strategy is applied to determine the noisy models best matching the actual mixture of speech and noisy background. The proposed algorithms are evaluated side by side with conventional compensation schemes on connected Czech digits presented in various levels of background car noise. The resulting system provides an absolute word error rate (WER) reduction on 10-dB signal-to-noise ratio data of 8.7% and 37.7% for female neutral and LE speech, respectively, and of 8.7% and 32.8% for male neutral and LE speech, respectively, when compared to the baseline recognizer employing perceptual linear prediction (PLP) coefficients and cepstral mean and variance normalization.Index Terms-Cepstral compensation, codebook of noisy models, frequency warping, Lombard effect, speech recognition.

show abstract

Section: Recognizer With Codebook Of Noisy Modelsmentioning

confidence: 99%

Section: Recognizer With Codebook Of Noisy Modelsmentioning

confidence: 99%

Unsupervised Equalization of Lombard Effect for Speech Recognition in Noisy Adverse Environments

Bořil

Hansen

2010

IEEE Trans. Audio Speech Lang. Process.

View full text Add to dashboard Cite

show abstract

“…These AR parameters were then used to provide car specific speech enhancement for noisy speech recognition. In another approach, Mokbel and Chollet (1995) proposed a feature enhancement/adaptation strategy where the characteristics of the Mel Frequency Cepstral Coefficients (MFCCs) in car environment were utilized to derive a non-linear Spectral Subtraction (SS) scheme. For perception centric systems, Arslan et al (1995) suggested adaptive noise suppression algorithms for mobile applications and tested them on simulated car noise.…”

Section: Speech Systems With Vehiclesmentioning

confidence: 99%

Car noise verification and applications

Krishnamurthy

Hansen

2013

Int J Speech Technol

View full text Add to dashboard Cite

This study presents audio based vehicle-verification as a new area of research. The task involves verifying the claim that an acoustic sample belongs to a vehicle. Audio based vehicle verification has the potential to impact research in the areas of vehicle forensics and in-vehicle speech systems. For this task, a new corpus (UTD-CAR-NOISE) that consists of noise from 20 vehicles under 8 distinct noise environments (∼8 hours of data). Our approach towards vehicle verification hypothesizes that some specific environments are more suited for vehicle verification. Towards this goal, four diverse in-vehicle noise conditions are identified on the basis of their frequency of occurrence. Additionally, four different verification systems are proposed based on their complexity and modeling strategies. Our evaluation shows that A/C on with windows closed condition is the most conducive for vehicle verification (98 %). The proposed systems were evaluated on approximately 100,000 trials, achieving performances in the range of (75-98 %) for different vehicle environments.

show abstract

“…By setting the singular values representing the n-oise to zero, we can reconstruct a new reduced rank matrix H representing the clean speech signal R using (4). To obtain the vector jz from H , the anti-diagonal components of H are then averaged.…”

Section: Singular Value Decomposition Based Speech Enhancementmentioning

confidence: 99%

“…Spectral subtraction has been shown to work to some extent for speech recognition [2] [4]. The spectral subtraction algorithm works on the assumption that the noise contained in the noisy speech signal is additive and uncorrelated with the clean speech signal.…”

Section: Introductionmentioning

confidence: 99%

Robust speech recognition using singular value decomposition based speech enhancement

Lilly

Paliwal

1996

The Journal of the Acoustical Society of America

View full text Add to dashboard Cite

Speech recognition systems work reasonably well in laboratory conditions, but their performance deteriorates drastically when they are deployed in practical situations where the speech is corrupted by additive noise. One way to improve the performance of a speech recognition system in the presence of noise, is to enhance the speech prior to its recognition. Two singular value decomposition based techniques have been recently proposed for speech enhancement [5] [6]. In these techniques, singular value decomposition has been applied to an over-determined, over-extended data matrix formed from the noisy speech signal. A noise-free, low rank approximation was obtained by retaining a specific number of singular values. This technique was applied here as a preprocessor for recognising speech in the presence of noise. It was found to improve the recognition performance significantly for signal-to-noise ratios less than 15dB.

show abstract

Automatic word recognition in cars

Cited by 45 publications

References 42 publications

Unsupervised Equalization of Lombard Effect for Speech Recognition in Noisy Adverse Environments

Unsupervised Equalization of Lombard Effect for Speech Recognition in Noisy Adverse Environments

Car noise verification and applications

Robust speech recognition using singular value decomposition based speech enhancement

Contact Info

Product

Resources

About