Parallel LVCSR Algorithm for Cellphone-Oriented Multicore Processors

Ishikawa, S.; Yamabana, Kiyoshi; Isotani, Ryosuke; Okumura, Akitoshi

doi:10.1109/icassp.2006.1659986

Cited by 16 publications

(8 citation statements)

References 6 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The implementation statically mapped a carefully partitioned recognition network onto the multiprocessors, but the 3.8 × speed up was limited by runtime load imbalance, which would not scale to 30+ multiprocessors. The authors of [10] explored coarse-grained concurrency in large vocabulary conversational speech recognition (LVCSR) and implemented a pipeline of tasks on a cellphone-oriented multicore architecture. [19] proposed a parallel LVCSR implementation on a commodity multicore system using OpenMP.…”

Section: Related Workmentioning

confidence: 99%

Parallelizing Speaker-Attributed Speech Recognition for Meeting Browsing

Friedland

Chong

Janin

2010

2010 IEEE International Symposium on Multimedia

View full text Add to dashboard Cite

The following article presents an application for browsing meeting recordings by speaker and keyword which we call the Meeting Diarist. The goal of the system is to enable browsing of the content with rich meta-data in a graphical user interface shortly after the end of meeting, even when the application runs on a contemporary laptop. We therefore developed novel parallel methods for speaker diarization and multi-hypothesis speech recognition that are optimized to run on multicore and manycore architectures. This paper presents the underlying parallel speaker diarization and speech recognition realizations, a comparison of results based on NIST RT07 evaluation data, and a description of the final application.

show abstract

Section: Related Workmentioning

confidence: 99%

Parallelizing Speaker-Attributed Speech Recognition for Meeting Browsing

Friedland

Chong

Janin

2010

2010 IEEE International Symposium on Multimedia

View full text Add to dashboard Cite

show abstract

“…Chong et al discusses opportunities that parallization of computationally intensive components can benefit the diarization applications [22]. Many speech and speaker recognition applications have already already seen performance benefits from parallelization [8], [9], [10], [11], [12] on parallel CPU and GPU platforms.…”

Section: Related Workmentioning

confidence: 99%

“…Modern GPUs can be programmed with languages such as CUDA [6] and OpenCL [7], which are intended to be used to speed up general-purpose and scientific applications. Many speech recognition and diarization applications have already seen tremendous speedups from being mapped onto parallel processors [8], [9], [10], [11], [12], [13], [14].…”

Section: Introductionmentioning

confidence: 99%

Fast speaker diarization using a high-level scripting language

Gonina

Friedland

Cook

et al. 2011

2011 IEEE Workshop on Automatic Speech Recognition &Amp; Understanding

View full text Add to dashboard Cite

Abstract-Current state-of-the-art speaker diarization systems use agglomerative clustering of Gaussian Mixture Models (GMMs) to determine the number of speakers in an audio recording. GMM training is a central computation in the agglomerative clustering approach, which presents computational challenges that limit performance and make real-time processing of audio very difficult. With the emergence of highly parallel multicore and manycore processors such as Graphics Processing Units (GPUs), we can re-implement GMM training for these processors to achieve faster than real-time performance by taking advantage of parallelism in the training computation. However, developing and maintaining the low-level GPU code is difficult and requires deep understanding of hardware architecture of the parallel processor. In this paper we present a speaker diarization application captured in under 50 lines of Python that achieves 50-200× faster than real-time performance by automatically executing computationally intensive GMM training on an NVIDIA GPU with no significant loss in accuracy.

show abstract

“…In [5], Ishikawa et al implemented a parallel speech recognition system in a cellphone using a 3-core processor. The system was divided in 3 steps, one for each core.…”

Section: Introductionmentioning

confidence: 99%

Using A* for the parallelization of speech recognition systems

Cardinal

Dumouchel

2012

2012 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)

View full text Add to dashboard Cite

The speed of modern processors has remained constant over the last few years but the integration capacity continues to follow Moore's law and thus, to be scalable, applications must be parallelized. This paper presents results in using the A* search algorithm in a large vocabulary speech recognition parallel system. This algorithm allows better parallelization over the Viterbi algorithm. First experiments with a "unigram approximation" heuristic resulted in approximatively 8.7 times less states being explored compared to our classical Viterbi decoder. The multi-thread implementation of the A* decoder led to a speed-up factor of 3 over its sequential counterpart .

show abstract

Parallel LVCSR Algorithm for Cellphone-Oriented Multicore Processors

Cited by 16 publications

References 6 publications

Parallelizing Speaker-Attributed Speech Recognition for Meeting Browsing

Parallelizing Speaker-Attributed Speech Recognition for Meeting Browsing

Fast speaker diarization using a high-level scripting language

Using A* for the parallelization of speech recognition systems

Contact Info

Product

Resources

About