Baidu USA LLC

Kexin Zhao

Wei Ping

Kainan Peng

Zhao Song

Mohammad Shoeybi

Markus Kliegl

Siddharth Goyal

Kavya Srinet

Described herein are systems and methods for compressing and speeding up dense matrix multiplications as found, for examples, in the fully connected and recurrent layers of neural networks for embedded large vocabulary continuous speech recognition (LVCSR). For compression, trace norm regularization technique embodiments were introduced and studied for training low rank factored versions of matrix multiplications. Compared to standard low rank training, the methods more consistently lead to good accuracy versus number of parameter trade-offs and can be used to speed-up training of large models. Faster inference may be further enabled on ARM processors through kernels optimized for small batch sizes, resulting in speed ups over the currently used library. Beyond LVCSR, the techniques are also generally applicable to embedded neural networks with large fully connected or recurrent layers.

Systems and methods for trace norm regularization and faster inference for embedded models

WaveFlow is a small-footprint generative flow for raw audio, which may be directly trained with maximum likelihood. WaveFlow handles the long-range structure of waveform with a dilated two-dimensional (2D) convolutional architecture, while modeling the local variations using expressive autoregressive functions. WaveFlow may provide a unified view of likelihood-based models for raw audio, including WaveNet and WaveGlow, which may be considered special cases. It generates high-fidelity speech, while synthesizing several orders of magnitude faster than existing systems since it uses only a few sequential steps to generate relatively long waveforms. WaveFlow significantly reduces the likelihood gap that has existed between autoregressive models and flow-based models for efficient synthesis. Its small footprint with 5.91M parameters makes it 15 times smaller than some existing models. WaveFlow can generate 22.05 kHz high-fidelity audio 42.6&#xd7; faster than real-time on a V100 graphics processing units (GPU) without using engineered inference kernels.

Small-footprint flow-based models for raw audio

Presented herein are embodiments of a non-autoregressive sequence-to-sequence model that converts text to an audio representation. Embodiment are fully convolutional, and a tested embodiment obtained about 46.7 times speed-up over a prior model at synthesis while maintaining comparable speech quality using a WaveNet vocoder. Interestingly, a tested embodiment also has fewer attention errors than the autoregressive model on challenging test sentences. In one or more embodiments, the first fully parallel neural text-to-speech system was built by applying the inverse autoregressive flow (IAF) as the parallel neural vocoder. System embodiments can synthesize speech from text through a single feed-forward pass. Also disclosed herein are embodiments of a novel approach to train the IAF from scratch as a generative model for raw waveform, which avoids the need for distillation from a separately trained WaveNet.

Parallel neural text-to-speech

Growing community of inventors

Santa Clara, CA, United States of America

Kexin Zhao