Blind Source Separation with Low Latency for In-Car Communication
Blind Source Separation with Low Latency for In-Car Communication
复制标题
DOI：
--
复制
发表时间：
2020
期刊：
影响因子：
--
通讯作者：
T. Ueda;S. Inoue;S. Makino;Mitsuo Matsumoto;Takeshi Yamada
中科院分区：
文献类型：
--
作者：
T. Ueda;S. Inoue;S. Makino;Mitsuo Matsumoto;Takeshi Yamada
我们研究了汽车低延迟盲源分离（BSS）。在频域BSS中，与短时傅里叶变换（STFT）相关的缓冲延迟是不可避免的。因此，我们需要缩短STFT帧长度。然而，由于瞬时混合模型的假设不成立，缩短STFT帧会降低源分离性能。在类似的研究中，最近提出了一种利用频域卷积混合模型构建高混响混合信号分离系统的方法。我们关注了该方法，并考虑了减少依赖于STFT帧的延迟的应用。我们在短混响条件下对该方法在汽车中的有效性进行了评估。实验结果表明，该方法不仅保持了原有的性能，而且减少了依赖于STFT帧的延迟。
We examined blind source separation (BSS) with low latency in cars. In frequency-domain BSS, a buffer delay associated with a short-time Fourier transform (STFT) is inevitable. Thus, we need to shorten the STFT frame length. However, shortening of the STFT frame degrades the source separation performance because an assumption of instantaneous mixture model does not hold. On a similar research, the method of formulating a system for separating highly reverberant mixture signals using a frequency-domain convolutive mixture model has recently proposed. We paid attention to the method and considered an application of reducing the delay dependent on the STFT frame. We evaluated the effectiveness of the method in cars under a short reverberant condition. Experimental results confirmed that the method not only maintained its performance but also reduced the delay dependent on the STFT frame.