TR2015-135

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


    •  Hori, T., Chen, Z., Erdogan, H., Hershey, J.R., Le Roux, J., Mitra, V., Watanabe, S., "The MERL/SRI System for the 3rd CHiME Challenge Using Beamforming, Robust Feature Extraction, and Advanced Speech Recognition", IEEE Workshop on Automatic Speech Recognition and Understanding (ASRU), DOI: 10.1109/​ASRU.2015.7404833, December 2015, pp. 475-481.
      BibTeX TR2015-135 PDF
      • @inproceedings{Hori2015dec2,
      • author = {Hori, T. and Chen, Z. and Erdogan, H. and Hershey, J.R. and {Le Roux}, J. and Mitra, V. and Watanabe, S.},
      • title = {The MERL/SRI System for the 3rd CHiME Challenge Using Beamforming, Robust Feature Extraction, and Advanced Speech Recognition},
      • booktitle = {IEEE Workshop on Automatic Speech Recognition and Understanding (ASRU)},
      • year = 2015,
      • pages = {475--481},
      • month = dec,
      • publisher = {IEEE},
      • doi = {10.1109/ASRU.2015.7404833},
      • url = {https://www.merl.com/publications/TR2015-135}
      • }
  • MERL Contact:
  • Research Areas:

    Artificial Intelligence, Speech & Audio

Abstract:

This paper introduces the MERL/SRI system designed for the 3rd CHiME speech separation and recognition challenge (CHiME-3). Our proposed system takes advantage of recurrent neural networks (RNNs) throughout the model from the front speech enhancement to the language modeling. Two different types of beamforming are used to combine multimicrophone signals to obtain a single higher quality signal. Beamformed signal is further processed by a single-channel bi-directional long short-term memory (LSTM) enhancement network which is used to extract stacked mel-frequency cepstral coefficients (MFCC) features. In addition, two proposed noise-robust feature extraction methods are used with the beamformed signal. The features are used for decoding in speech recognition systems with deep neural network (DNN) based acoustic models and large-scale RNN language models to achieve high recognition accuracy in noisy environments. Our training methodology includes data augmentation and speaker adaptive training, whereas at test time model combination is used to improve generalization. Results on the CHiME-3 benchmark show that the full cadre of techniques substantially reduced the word error rate (WER). Combining hypotheses from different robust-feature systems ultimately achieved 9.10% WER for the real test data, a 72.4% reduction relative to the baseline of 32.99% WER.

 

  • Related News & Events

    •  AWARD    MERL's Speech Team Achieves World's 2nd Best Performance at the Third CHiME Speech Separation and Recognition Challenge
      Date: December 15, 2015
      Awarded to: John R. Hershey, Takaaki Hori, Jonathan Le Roux and Shinji Watanabe
      MERL Contact: Jonathan Le Roux
      Research Area: Speech & Audio
      Brief
      • The results of the third 'CHiME' Speech Separation and Recognition Challenge were publicly announced on December 15 at the IEEE Automatic Speech Recognition and Understanding Workshop (ASRU 2015) held in Scottsdale, Arizona, USA. MERL's Speech and Audio Team, in collaboration with SRI, ranked 2nd out of 26 teams from Europe, Asia and the US. The task this year was to recognize speech recorded using a tablet in real environments such as cafes, buses, or busy streets. Due to the high levels of noise and the distance from the speaker's mouth to the microphones, this is very challenging task, where the baseline system only achieved 33.4% word error rate. The MERL/SRI system featured state-of-the-art techniques including multi-channel front-end, noise-robust feature extraction, and deep learning for speech enhancement, acoustic modeling, and language modeling, leading to a dramatic 73% reduction in word error rate, down to 9.1%. The core of the system has since been released as a new official challenge baseline for the community to use.
    •