Parallel Architectures for Speech Recognition: Putting It All Together

这项支持用于高级语音识别算法体系结构的研究，以及布朗大学工程人机系统实验室(LEMS)的实验。作为研究设施的基础，研究人员使用了Armstrong，一个80节点松散耦合的分布式多处理器系统，以及一个由线性麦克风阵列组成的硬件/软件试验台，用于语音采集。所解决的问题包括语音分析的时间相关建模的计算效率算法的研究，连接语音的实时早期决策识别的新算法的研究，以及对高度时变的语音事件的隐马尔可夫和神经网络表示的精细时间分辨率的研究。

This support is for research on advanced speech recognition algorithms architectures, and experimentation in the Laboratory for Engineering Man/Machine Systems (LEMS) at Brown University. As a basis for the facility of research, the researchers use Armstrong, an 80-node loosely coupled distributed multiprocessor system, and a hardware/software testbed composed of a linear array of microphones for speech acquisition. Problems addressed include the investigation of computationally efficient algorithms for time-dependent modeling of speech analysis, study on new algorithms for real-time early-decision recognition of connected speech, and research on fine-time resolution of hidden Markov and neural-network representations of highly time-varying phonic events.