LINEAR PREDICTION-BASED DEREVERBERATION WITH ADVANCED SPEECH ENHANCEMENT AND RECOGNITION TECHNOLOGIES FOR THE REVERB CHALLENGE

By Marc Delcroix, Takuya Yoshioka, Atsunori Ogawa, Yotaro Kubo, Masakiyo Fujimoto, Nobutaka Ito, Keisuke Kinoshita, Miquel Espi, Takaaki Hori, Tomohiro Nakatani and 1 more. 2014

112 citations recorded by the source corpus.

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

1. Jinyu Li
2. Shinji Watanabe
3. Bo Li
4. Tara N. Sainath
5. Mirco Ravanelli
6. Dong Yu
7. Wei-Ning Hsu
8. Yu Zhang

Citations in the corpus

• Speech dereverberation and denoising using complex ratio masks (2017, background): “Although complex domain dereverberation is presented in [16, 17], their approach is unsupervised and does not handle background noise.”
• Speech Dereverberation Using Fully Convolutional Networks (2018, background): “In [4], the signal is dereverberated by estimating the inverse filter that minimizes the weighted linear prediction error (WPE).”
• NARA-WPE: A Python package for weighted prediction error dereverberation in Numpy and Tensorflow for online and offline processing (2018, background): “High ASR performance gains were presented during the REVERB challenge [8–10] where WPE was able to improve results on top of an already competitive beamforming frontend [11].”
• Speech acoustic modeling from raw multichannel waveforms (2015, background): “[8] who combine a speech enhancement frontend comprised of linear prediction derever­ beration, beamforming, and model-based noise reduction at the input to an ASR system.”
• JHU ASpIRE system: Robust LVCSR with TDNNS, iVector adaptation and RNN-LMS (2015, methodology): “Alignments for clean speech were generated using the GMM-HMM system, using clean data as recommended in [23, 24].”
• An Investigation into the Effectiveness of Enhancement in ASR Training and Test for Chime-5 Dinner Party Transcription (2019, background): “Also, dereverberation algorithms have been shown to consistently improve ASR results, since the temporal dispersion of the signal caused by reverberation is difficult to capture by an ASR acoustic model [1].”
• Spatial diffuseness features for DNN-based speech recognition in noisy and reverberant environments (2014, background): “, in the form of beamforming [4], multichannel linear prediction [5], blocking matrix-based postfilters [6] or coherence-based postfilters [7].”
• Multichannel Signal Processing With Deep Neural Networks for Automatic Speech Recognition (2017, methodology): “We compare to results for 8 channel oracle D+S, where the true target speech TDOA is known, and to oracle MVDR [6] where the true noise covariance is known in addition to the target TDOA.”
• The NTT CHiME-3 system: Advances in speech enhancement and recognition for mobile multi-microphone devices (2015, methodology): “Top-performing systems in recent challenge programs associated with noise robustness integrate strong front-ends and state-of-the-art back-ends [1, 2, 3].”