The NTT CHiME-3 system: Advances in speech enhancement and recognition for mobile multi-microphone devices

By Takuya Yoshioka, Nobutaka Ito, Marc Delcroix, Atsunori Ogawa, Keisuke Kinoshita, Masakiyo Fujimoto, Chengzhu Yu, Wojciech J. Fabian, Miquel Espi, Takuya Higuchi and 2 more. 2015

250 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Speech & Audio, Signal Processing

How researchers see this paper

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

In Artificial Intelligence

In Speech & Audio

In Signal Processing

Citations in the corpus

Toward Domain-Invariant Speech Recognition via Large Scale Training (2018, background): “Specialized techniques like masking [9] and beamforming [10] (in the case of multi-microphone input) help in certain conditions.”
Deep Speech 2 : End-to-End Speech Recognition in English and Mandarin (2015, background): “Using all 6 channels of the CHiME audio can provide substantial performance improvements (Yoshioka et al., 2015).”
The third ‘CHiME’ speech separation and recognition challenge: Dataset, task and baselines (2015, unknown): “Other teams trained systems on all the individual channels rather than on a single enhanced signal [18, 22, 24, 30].”
Batch-normalized joint training for DNN-based distant speech recognition (2016, background): “To counteract such adversities, even the most recent DSR systems [9] must rely on a combination of several interconnected technologies, including for instance speech enhancement [10], speech separation [11], acoustic event detection and classiﬁcation [12, 13], speaker identiﬁcation [14], speaker…”
A Review of Speaker Diarization: Recent Advances with Deep Learning (2021, background): “However, recent advances in the areas of Speech Recognition [200, 201], far-field speech processing [202, 203, 204], Speaker ID and diarization [205, 206, 113],”
Student-teacher network learning with enhanced features (2017, background): “In particular, when using multichannel speech enhancement techniques such as beamforming [5], significant improvements can be obtained compared to a single-channel ASR systems, and the state-of-the-art multichannel systems are approaching the performance of ASR systems in clean conditions [6, 7].”
Improved MVDR Beamforming Using Single-Channel Mask Prediction Networks (2016, background): “Predicting time-frequency masks and utilizing them for beamforming have been considered in other recent studies as well [16, 17, 18, 19].”
The RWTH/UPB/FORTH System Combination for the 4th CHiME Challenge Evaluation (2016, methodology): “For the training of the model L2, the sentences with high OOV rates were removed from the training data, exactly as described in [24].”
Realistic Multi-Microphone Data Simulation for Distant Speech Recognition (2016, background): “Approaches as contaminated speech training [14, 15, 16], multi-style training [17, 18, 19] and data augmentation [20, 21, 22, 23] have, in fact, been shown very effective in improving the DSR system performance.”
Deep Learning in Mobile and Wireless Networking: A Survey (2018, background): “[250] Mobile speech recognition Cloud-based CNN Ruan et al.”
The third 'CHiME' speech separation and recognition challenge: Analysis and outcomes (2017, unknown): “Alternatives to the baseline DNN have included convolutional neural net-workss (CNNs), (e.g., Yoshioka et al., 2015; Wang et al., 2015; Zhuang et al., 2015; Ma et al., 2015; Baby et al., 2015) and forms of Long Short Time Memory (LSTM) networks, (e.g., Du et al., 2015; Wang et al., 2015; Zhuang et…”
A network of deep neural networks for Distant Speech Recognition (2017, methodology): “To counteract such adversities, most DSR systems must rely on a combination of several interconnected technologies [6], including methods for speech enhancement [7], speech separation [8], acoustic event classification [9, 10], speaker identification [11], just to name a few.”
Deep Learning for Environmentally Robust Speech Recognition (2017, methodology): “orming, the winning contribution was still able to rely on the baseline MVDR beamformer complemented with an improved DOA estimate (the steering vector) based on T-F masks estimated using complex GMM [125], [133]. Recent studies have, however, started to investigate the use of DNN in the earliest st”
Recent progresses in deep learning based acoustic models (2017, methodology): “The current dominant approach is still using the traditional beamforming method to process the waveforms from multiple microphones, and then inputting the beamformed signal into the acoustic model [56].”
Domain adaptation of DNN acoustic models using knowledge distillation (2017, methodology): “The source (baseline) model was a convolutional neural network with network-in-network architecture (NiN-CNN) acoustic model [20].”