The MERL/SRI system for the 3RD CHiME challenge using beamforming, robust feature extraction, and advanced speech recognition

An analysis of environment, microphone and data simulation mismatches in robust speech recognition

“…Concerning other parts of the decoder, Hori et al (2015) reported consistent improvements on real and simulated data by replacing the default 3-gram language model used in the baseline by a 5-gram language model with Kneser-Ney (KN) smoothing (Kneser and Ney, 1995), rescoring the lattice using a recurrent neural network language model (RNN-LM) (Mikolov et al, 2010), and fusing the outputs of multiple systems using MBR. This claim also holds true for system combination based on recognizer output voting error reduction (ROVER) (Fiscus, 1997), as reported by Fujita et al (2015).…”

“…In the following, we do not discuss DNN post-filters, which provided a limited improvement or degradation on both real and simulated data (Hori et al, 2015;Sivasankaran et al, 2015), and we focus on multichannel DNN-based enhancement instead. Table 5 illustrates the performance of the DNN-based time-invariant generalized eigenvalue (GEV) beamformer proposed by Heymann et al (2015).…”

“…Each training setup resulted in a different enhancement system. The resulting ASR performance was evaluated using the updated DNNbased baseline distributed by the organizers after the challenge 4 (Hori et al, 2015). This baseline is identical to the one described in Section 3.1, except that decoding is performed using a 5-gram language model with KN smoothing and RNN-LM based rescoring.…”

“…Other enhancement techniques which result in similar characteristics for enhanced real and simulated signals do not appear to suffer from this problem. Table 7: WER (%) achieved after enhancement by BeamformIt using various feature extraction and normalization methods and the DNN backend retrained on enhanced real and simulated data without sMBR (Hori et al, 2015 Tachioka et al (2015) concatenated logmel or MFCC features with 40-dimensional bottleneck (BN) features extracted as the neuron outputs in the smaller hidden layer of a neural network with two hidden layers trained to predict phoneme posteriors. The neural network was trained on real and simulated data with logmel and pitch features as inputs.…”

The third ‘CHiME’ speech separation and recognition challenge: Analysis and outcomes

“…Heymann et al (2015) employ a DNN to perform the necessary speech and noise covariance estimates. Other teams have employed a conventional delay and sum beamformer (e.g., Sivasankaran et al, 2015;Hori et al, 2015;Prudnikov et al, 2015). Of these, several reported that the freely available BeamformIt tool developed by Anguera et al (2007) worked very effectively.…”