Automatic gain control and multi-style training for robust small-footprint keyword spotting with deep neural networks

By Rohit Prabhavalkar, Raziel Alvarez, Carolina Parada, Preetum Nakkiran, Tara N. Sainath. 2015

93 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

1. Tara N. Sainath
2. Jinyu Li
3. Rohit Prabhavalkar
4. Dong Yu
5. Quan Wang
6. Ignacio Lopez-Moreno
7. Quoc V. Le
8. Yonghui Wu

In Speech & Audio

1. Tara N. Sainath
2. Quan Wang
3. Ignacio Lopez-Moreno
4. Jinyu Li
5. Li Wan
6. Rohit Prabhavalkar
7. Raziel Alvarez
8. Yanzhang He

In Signal Processing

1. Rohit Prabhavalkar
2. Tara N. Sainath
3. Jinyu Li
4. Quan Wang
5. Ignacio Lopez-Moreno
6. Ian McGraw
7. Yanzhang He
8. Li Wan

Citations in the corpus

• SpecAugment: A Simple Data Augmentation Method for Automatic Speech Recognition (2019, neutral): “Data augmentation for keyword spotting has been studied in [15, 16].”
• Generalized End-to-End Loss for Speaker Verification (2017, methodology): “For multi-user on Google Home, two keywords are supported simultaneously: “OK Google” and “Hey Google”.”
• End-to-end text-dependent speaker verification (2015, methodology): “6 seconds long global password relates to the Google Keyword Spotting system [2] and Google VoiceSearch [3] and facilitates the combination of the systems.”
• Locally-connected and convolutional neural networks for small footprint speaker recognition (2015, methodology): “6 seconds long global password was chosen as it relates to the Google Keyword Spotting system [2] and Google VoiceSearch [3], facilitating the combination of all three systems.”
• Training Keyword Spotting Models on Non-IID Data with Federated Learning (2020, background): “Based on a priori distributions, MTR generates random room sizes and dimensions, speaker and noise source positions, signal to noise ratios, and reverberation times [39].”
• Compressing deep neural networks using a rank-constrained topology (2015, methodology): “Our systems are trained on multi-style training data created by adding car and cafeteria noise to the clean training data at an SNR in the range [-5dB, +10dB] as detailed in [10].”
• 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].”
• Attention-Based Models for Text-Dependent Speaker Verification (2017, methodology): “The same acoustic features are used for both keyword detection [10] and speaker verification.”
• Speaker Diarization with LSTM (2017, background): “However, most of these applications belong to text-dependent speaker verification, where the speaker embeddings are extracted from specific detected keywords [12, 13].”