Real-Time Blind Speaker Extraction

Microphone-array signals processed by blockwise ILRMA and framewise RCSCME to extract target speech

We develop low-latency multichannel blind speech extraction for real acoustic environments, focusing on directional speech embedded in diffuse noise. The methods combine spatially regularized independent low-rank matrix analysis (ILRMA) with rank-constrained spatial covariance matrix estimation (RCSCME), and investigate fast demixing updates, blockwise and online processing, and asymmetric analysis windows for real-time operation. This research line also includes earlier probabilistic multichannel blind source separation, including DOA-HMM-based moving-source tracking and factorial-HMM-based joint source separation and activity detection; these earlier methods are not presented as real-time systems.

Related Publications

Yuto Ishikawa, Norihiro Takamune, Kouei Yamaoka, Tomohiko Nakamura, and Hiroshi Saruwatari, “Joint optimization of demixing filters and asymmetric window function for independent vector analysis,” in Proceedings of International Workshop on Acoustic Signal Enhancement, Sept. 2026.
Yuto Ishikawa, Tomohiko Nakamura, Norihiro Takamune, Daichi Kitamura, Hiroshi Saruwatari, Yu Takahashi, and Kazunobu Kondo, “Low-latency real-time speech extraction based on rank-constrained spatial covariance matrix estimation using asymmetric window function,” in Joint Meeting of the Acoustical Society of America and the Acoustical Society of Japan, Dec. 2025.
Yuto Ishikawa, Tomohiko Nakamura, Norihiro Takamune, Daichi Kitamura, Hiroshi Saruwatari, Yu Takahashi, and Kazunobu Kondo, “Real-time speech extraction based on rank-constrained spatial covariance matrix estimation and spatially regularized independent low-rank matrix analysis with fast demixing matrix estimation,” IEEE Access, vol. 13, pp. 88683–88706, May 2025.
Yuto Ishikawa, Tomohiko Nakamura, Norihiro Takamune, and Hiroshi Saruwatari, “Hearing-aids system using distributed assistive device and blind speech extraction method under diffuse noise,” in International Congress on Acoustics, May 2025.
Yuto Ishikawa, Tomohiko Nakamura, Norihiro Takamune, and Hiroshi Saruwatari, “Real-time framework for speech extraction based on independent low-rank matrix analysis with spatial regularization and rank-constrained spatial covariance matrix estimation,” in Workshop on Spoken Dialogue Systems for Cybernetic Avatars (SDS4CA), Sept. 2024.
Yuto Ishikawa, Kohei Konaka, Tomohiko Nakamura, Norihiro Takamune, and Hiroshi Saruwatari, “Real-time speech extraction using spatially regularized independent low-rank matrix analysis and rank-constrained spatial covariance matrix estimation,” in Proceedings of Hands-Free Speech Communication and Microphone Arrays, IEEE International Conference on Acoustics, Speech, and Signal Processing Workshops, Apr. 2024, pp. 730–734.
Takuya Higuchi, Hirofumi Takeda, Tomohiko Nakamura, and Hirokazu Kameoka, “A unified approach for underdetermined blind signal separation and source activity detection by multichannel factorial hidden Markov models,” in Proceedings of INTERSPEECH, Sept. 2014, pp. 850–854.
Takuya Higuchi, Norihiro Takamune, Tomohiko Nakamura, and Hirokazu Kameoka, “Underdetermined blind separation and tracking of moving sources based on DOA-HMM,” in Proceedings of IEEE International Conference on Acoustics, Speech, and Signal Processing, May 2014, pp. 3215–3219.