Small-footprint keyword spotting using deep neural networks

By Guoguo Chen, Carolina Parada, Georg Heigold. 2014

589 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Speech & Audio

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

Citations in the corpus

Federated Learning for Keyword Spotting (2018, methodology): “A posterior handling [16] generates a conﬁdence score for every frame by combining the smoothed label posteriors.”
Structured Transforms for Small-Footprint Deep Learning (2015, methodology): “Speciﬁcally, we consider a keyword spotting (KWS) task, where a deep neural network is trained to detect a speciﬁc phrase, such as “Ok Google” [2].”
Quaternion Recurrent Neural Networks (2018, background): “This characteristic makes our QRNN model particularly suitable for speech recognition conducted on low computational power devices like smartphones and tablets [6].”
Streaming small-footprint keyword spotting using sequence-to-sequence models (2017, background): “Many previous works propose to train neural networks to identify word targets in individual keywords: for example, using feed-forward deep neural networks [4, 5, 6], convolutional networks [7] or recurrent neural networks [8, 9, 10].”
Wav2KWS: Transfer Learning From Speech Representations for Keyword Spotting (2021, background): “Keyword spotting (KWS) recognizes utterances of a limited number of commands and is being actively studied along with ASR that recognizes continuous speech [10]–[14].”
Survey on Deep Neural Networks in Speech and Vision Systems (2019, methodology): “Several studies [167-170, 173] have customized DNN, CNN, and recurrent LSTM architectures with compression and quantization schemes to achieve considerable reductions in memory and computational requirements.”
Segmental acoustic indexing for zero resource keyword search (2015, background): “Query-by-example search, where search terms are presented as audio segments rather than in graphemic or phonetic form, has applications in probing large collections of unstructured audio data [4] and in voice interfaces [5].”
DeepX: A Software Accelerator for Low-Power Deep Learning Inference on Mobile Devices (2016, background): “We use one CNN and two DNNs of this type.”
Machine Learning at the Network Edge: A Survey (2019, methodology): “maller networksize,anda16timessmallermemoryfootprintthanstate-of-the-artDNNs. Chenetal. introducedasmall-footprintkeywordspottingtechniquebasedonDNNscalledDeepKWS, whichissuitableformobileedge-devices[23]. DeepKWShasthreecomponents: afeatureextractor,adeep neural network, and a posterior handling mo”
Training Keyword Spotting Models on Non-IID Data with Federated Learning (2020, background): “Applications of neural networks to keyword spotting have also been explored, particularly within the contexts of quality improvement and latency reduction for low-resource environments [4, 5, 6, 7, 8, 9] and end-to-end model training [10].”
Efficient Keyword Spotting Using Dilated Convolutions and Gating (2018, methodology): “End-to-end models such as the one presented here do not require any postprocessing step besides smoothing, as opposed to multi-class models such as [4, 5].”
Compressing deep neural networks using a rank-constrained topology (2015, methodology): “Our baseline system (baseline-240K) is based on the system in [9], which consists of a DNN trained to predict individSystem Initialization Total Parameters”
TensorFlow Lite Micro: Embedded Machine Learning on TinyML Systems (2020, background): “The most well-known and widely deployed example of this new TinyML technology is keyword spotting, also called hotword or wakeword detection (Chen et al., 2014; Gru-enstein et al., 2017; Zhang et al., 2017).”
Streaming keyword spotting on mobile devices (2020, methodology): “The standard end to end model architecture applied for KWS [12, 14] consists of a speech feature extractor(optional) and a neural network based classifier.”
Layer-wise training of deep networks using kernel similarity (2017, background): “ems such as object detection [4], English character recognition [10]. It also showed promising results for speech data where it has been applied for speech recognition [6] and spoken keyword spotting [5]. It has been studied that the effectiveness of deep networks lies in the layered representation ”
Contaminated speech training methods for robust DNN-HMM distant speech recognition (2017, background): “Note also that the use of an ACW does not imply any additional computational effort and is particularly suitable for on-line systems [34, 35], since the latency originated by waiting for the future frames is minimized.”