QuartzNet: Deep Automatic Speech Recognition with 1D Time-Channel Separable Convolutions

2019/10/22 by Kriman, Samuel, Beliaev, Stanislav, Ginsburg, Boris +6 · 4 citations
#Audio and Speech Processing (eess.AS) #FOS: Electrical engineering #electronic engineering #information engineering

paper · doi:10.48550/arxiv.1910.10261

Abstract

We propose a new end-to-end neural acoustic model for automatic speech recognition. The model is composed of multiple blocks with residual connections between them. Each block consists of one or more modules with 1D time-channel separable convolutional layers, batch normalization, and ReLU layers. It is trained with CTC loss. The proposed network achieves near state-of-the-art accuracy on LibriSpeech and Wall Street Journal, while having fewer parameters than all competing models. We also demonstrate that this model can be effectively fine-tuned on new datasets.

Cited by

• Real-Time Streamable Generative Speech Restoration with Flow Matching
• Dealing with the Hard Facts of Low-Resource African NLP
• Scriboora: Rethinking Human Pose Forecasting
• Are These Even Words? Quantifying the Gibberishness of Generative Speech Models

Related

• Citrinet: Closing the Gap between Non-Autoregressive and Autoregressive End-to-End Models for Automatic Speech Recognition
• Multi-QuartzNet: Multi-Resolution Convolution for Speech Recognition with Multi-Layer Feature Fusion
• Residual Convolutional CTC Networks for Automatic Speech Recognition
• SpeakerNet: 1D Depth-wise Separable Convolutional Network for Text-Independent Speaker Recognition and Verification
• Densely Connected Convolutional Networks for Speech Recognition