Speech Commands: A Dataset for Limited-Vocabulary Speech Recognition

By Pete Warden. 2018

2,040 citations recorded by the source corpus.

Describes an audio dataset of spoken words designed to help train and evaluate keyword spotting systems. Discusses why this task is an interesting challenge, and why it requires a specialized dataset that is different from conventional datasets used for automatic speech recognition of full sentences. Suggests a methodology for reproducible and comparable accuracy metrics for this task. Describes how the data was collected and verified, what it contains, previous versions and properties. Concludes by reporting baseline results of models trained on this dataset.

Subfields: Artificial Intelligence, Evaluation & Benchmarks, 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. Yoshua Bengio
2. Quoc V. Le
3. Mohit Bansal
4. James R. Glass
5. Jianfeng Gao
6. Percy Liang
7. Luke S. Zettlemoyer
8. Furu Wei

In Evaluation & Benchmarks

1. Yoshua Bengio
2. James R. Glass
3. Herman Kamper
4. Ali Farhadi
5. Jonathan Berant
6. Song Han
7. Yuan Gong
8. Honglak Lee

In Signal Processing

1. James R. Glass
2. Hung-yi Lee
3. Jinyu Li
4. Wei-Ning Hsu
5. Yu-An Chung
6. Dong Yu
7. Yoshua Bengio
8. Shinji Watanabe

Citations in the corpus

• On Fast Sampling of Diffusion Probabilistic Models (2021, methodology): “We conduct unconditional and class-conditional audio synthesis experiments on the Speech Commands 0-9 (SC09) dataset, the spoken digit subset of the full Speech Commands dataset (Warden, 2018).”
• DiffWave: A Versatile Diffusion Model for Audio Synthesis (2020, methodology): “Data: We use the Speech Commands dataset (Warden, 2018), which contains many spoken words by thousands of speakers under various recording conditions including very noisy environment.”
• BigSSL: Exploring the Frontier of Large-Scale Semi-Supervised Learning for Automatic Speech Recognition (2021, neutral): “8 Speech Commands [93] Command 12 100,503 1.”
• Federated Learning for Keyword Spotting (2018, background): “The reference dataset for multi-class keyword spotting is the speech command dataset [13], but the speech command task is generally preceded by a wake word detector and is focused on minimizing the confusion across classes, not robustness to false alarms.”
• Federated Learning with Non-IID Data (2018, methodology): “1 Experimental Setup In this work, we use convolutional neural networks (CNNs) trained on MNIST [6], CIFAR-10 [7] and Speech commands datasets [29].”
• Resource-Efficient Neural Architect (2018, methodology): “For KWS task, we use the Google speech commands dataset [Warden, 2018].”
• FrugalML: How to Use ML Prediction APIs More Accurately and Cheaply (2020, neutral): “Dataset Size # Classes Dataset Size # Classes Tasks FER+ [20] 6358 7 RAFDB [39] 15339 7 FER EXPW [59] 31510 7 AFFECTNET [42] 287401 7 YELP [18] 20000 2 SHOP [15] 62774 2 SA IMDB [41] 25000 2 WAIMAI [17] 11987 2 DIGIT [5] 2000 10 AUDIOMNIST [21] 30000 10 STT FLUENT [40] 30043 31 COMMAND [52] 64727 31”
• Soft Threshold Weight Reparameterization for Learnable Sparsity (2020, methodology): “(Warden, 2018).”
• Speech Model Pre-training for End-to-End Spoken Language Understanding (2019, positive): “The Google Speech Commands dataset [29] (to which the name “Fluent Speech Commands” is an homage) is a free dataset of 30 single-word spoken commands (“yes”, “no”, “stop”, “go”, etc.”
• Neural Controlled Differential Equations for Irregular Time Series (2020, neutral): “We used the Speech Commands dataset [34].”
• AST: Audio Spectrogram Transformer (2021, methodology): “We then present our experiments on ESC-50 and Speech Commands V2 in Section 3.2.”
• GAN You Hear Me? Reclaiming Unconditional Speech Synthesis from Diffusion Models (2022, methodology): “To compare to existing unconditional speech synthesis models, we use the Google Speech Commands dataset of isolated spoken words [17].”
• Detecting AI Trojans Using Meta Neural Analysis (2019, methodology): “We use the SpeechCommand dataset (SC) version 0.02 [52] for the speech task.”
• AudioCaps: Generating Captions for Audios in The Wild (2019, background): “Speech Commands dataset (Warden, 2018), Common Voice dataset (Mozilla, 2017), Librispeech (Panayotov et al., 2015), LS Speech (Ito, 2017).”
• MCUNet: Tiny Deep Learning on IoT Devices (2020, methodology): “We used 3 datasets as benchmark: ImageNet [14], Visual Wake Words (VWW) [12], and Speech Commands (V2) [44].”