Vector quantization for the efficient computation of continuous density likelihoods

Bocchieri, Enrico

doi:10.1109/icassp.1993.319405

Cited by 120 publications

(81 citation statements)

References 3 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…If we could combine them efficiently, further process acceleration could be expected. Based on the similar idea, the combination of Gaussian selection [4] and batch calculation [10] is proposed in [12], and the complemental process acceleration is reported. However, to the best of our knowledge, there have been no studies investigating the combination of recycling [5], [6] and batch calculation [10].…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Efficient Combination of Likelihood Recycling and Batch Calculation for Fast Acoustic Likelihood Calculation

Ogawa

Takahashi

Nakamura

2011

IEICE Trans. Inf. & Syst.

View full text Add to dashboard Cite

SUMMARYThis paper proposes an efficient combination of state likelihood recycling and batch state likelihood calculation for accelerating acoustic likelihood calculation in an HMM-based speech recognizer. Recycling and batch calculation are each based on different technical approaches, i.e. the former is a purely algorithmic technique while the latter fully exploits computer architecture. To accelerate the recognition process further by combining them efficiently, we introduce conditional fast processing and acoustic backing-off. Conditional fast processing is based on two criteria. The first potential activity criterion is used to control not only the recycling of state likelihoods at the current frame but also the precalculation of state likelihoods for several succeeding frames. The second reliability criterion and acoustic backing-off are used to control the choice of recycled or batch calculated state likelihoods when they are contradictory in the combination and to prevent word accuracies from degrading. Large vocabulary spontaneous speech recognition experiments using four different CPU machines under two environmental conditions showed that, compared with the baseline recognizer, recycling and batch calculation, our combined acceleration technique further reduced both of the acoustic likelihood calculation time and the total recognition time. We also performed detailed analyses to reveal each technique's acceleration and environmental dependency mechanisms by classifying types of state likelihoods and counting each of them.

show abstract

Section: Introductionmentioning

confidence: 99%

“…The first category consists of purely algorithmic techniques, such as Gaussian reduction [1], model parameter tying [2], scalar quantization of feature vectors [3], Gaussian selection [4], and state selection (or state likelihood recycling) [5], [6]. All of these techniques are based on approx- imations, i.e.…”

Section: Introductionmentioning

confidence: 99%

Efficient Combination of Likelihood Recycling and Batch Calculation for Fast Acoustic Likelihood Calculation

Ogawa

Takahashi

Nakamura

2011

IEICE Trans. Inf. & Syst.

View full text Add to dashboard Cite

show abstract

“…For small and medium vocabulary tasks, the state likelihood computation can represent a significant portion of the overall computation. One way to speed up this computation is to reduce the number of Gaussians needing to be considered to compute the likelihood for a state by preparing a Gaussian short list for each HMM state and each region of the quantified feature space [5]. Doing so, only a fraction of the Gaussians of each mixture is considered during decoding.…”

Section: Decodermentioning

confidence: 99%

Speech recognition for an information kiosk

Gauvain

Gangolf

Lamel

Proceeding of Fourth International Conference on Spoken Language Processing. ICSLP '96

View full text Add to dashboard Cite

In the context of the ESPRIT MASK project we face the problem of adapting a "state-of-the-art" laboratory speech recognizer for use in the real world with naive users. The speech recognizer is a softwareonly system that runs in real-time on a standard Risc processor. All aspects of the speech recognizer have been reconsidered from signal capture to adaptive acoustic models and language models. The resulting system includes such features as microphone selection, response cancellation, noise compensation, query rejection capability and decoding strategies for real-time recognition.

show abstract

“…Another way to reduce the search area is to cluster the Gaussians after the training and using the obtained cluster centers to determine the exact search area [2]. This can also be done during the training by organizing the Gaussians initially into a tree structure and maintaining the topological ordering in the structure by training the models with the Tree-search SOM [6].…”

Section: Introductionmentioning

confidence: 99%

“…A remarkable part of the whole recognition time (clearly over 50% for the current system) is consumed in the search for the best-matching mixtures and for larger systems the portion increases even more. With large mixture density codebooks it is often a sufficient density approximation to compute only a few mixtures (Kbest) closest to the current observation, because the other mixtures will produce such a low and inaccurate likelihood values that they have practically no effect on the search for the optimal path [7,1,2].…”

Section: Introductionmentioning

confidence: 99%

Using the self-organizing map to speed up the probability density estimation for speech recognition with mixture density HMMs

Kurimo¹,

Somervuo²

Proceeding of Fourth International Conference on Spoken Language Processing. ICSLP '96

View full text Add to dashboard Cite

This paper presents methods to improve the probability density estimation in hidden Markov models for phoneme recognition by exploiting the Self-Organizing Map (SOM) algorithm. The advantage of using the SOM is based on the created approximative topology between the mixture densities by training the Gaussian mean vectors used as the kernel centers by the SOM algorithm. The topology makes the neighboring mixtures to respond strongly for the same inputs and so most of the nearest mixtures used to approximate the current observation probability will be found in the topological neighborhood of the "winner" mixture. Also the knowledge about the previous winners are used to speed up the the search for the new winners. Tree-search SOMs and segmental SOM training are studied aiming at faster search and suitability for HMM training. The framework for the presented experiments includes melcepstrum features and phoneme-wise tied mixture density HMMs.

show abstract

Vector quantization for the efficient computation of continuous density likelihoods

Cited by 120 publications

References 3 publications

Efficient Combination of Likelihood Recycling and Batch Calculation for Fast Acoustic Likelihood Calculation

Efficient Combination of Likelihood Recycling and Batch Calculation for Fast Acoustic Likelihood Calculation

Speech recognition for an information kiosk

Using the self-organizing map to speed up the probability density estimation for speech recognition with mixture density HMMs

Contact Info

Product

Resources

About