Toward Domain-Invariant Speech Recognition via Large Scale Training

Abstract

Current state-of-the-art automatic speech recognition systems are trained to work in specific 'domains', defined based on factors like application, sampling rate and codec. When such recognizers are used in conditions that do not match the training domain, performance significantly drops. This work explores the idea of building a single domain-invariant model for varied use-cases by combining large scale training data from multiple application domains. Our final system is trained using 162,000 hours of speech. Additionally, each utterance is artificially distorted during training to simulate effects like background noise, codec distortion, and sampling rates. Our results show that, even at such a scale, a model thus trained works almost as well as those fine-tuned to specific subsets: A single model can be robust to multiple application domains, and variations like codecs and noise. More importantly, such models generalize better to unseen conditions and allow for rapid adaptation -we show that by using as little as 10 hours of data from a new domain, an adapted domain-invariant model can match performance of a domain-specific model trained from scratch using 70 times as much data. We also highlight some of the limitations of such models and areas that need addressing in future work.

Deliberation Model Based Two-Pass End-To-End Speech Recognition

“…For training, we use the same multidomain datasets as in [20,21] which include anonymized and hand-transcribed English utterances from general Google traffic, far-field environments, telephony conversations, and YouTube. We augment the clean training utterances by artificially corrupting them by using a room simulator, varying degrees of noise, and reverberation such that the signal-to-noise ratio (SNR) is between 0dB and 30dB [23].…”

“…Our experiments are conducted using the same training data as in [20,21], which is from multiple domains such as Voice Search, YouTube, Farfield and Telephony. We first analyze the behavior of the deliberation model, including performance when attending to multiple RNN-T hypotheses, contribution of different attention, and rescoring vs. beam search.…”

End-to-End Training of a Large Vocabulary End-to-End Speech Recognition System

“…Another important aspect in building high-performance speech recognition systems is the amount and the coverage of the training data. To build high performance speech recognition systems for conversational speech, we need to use a large amount of speech data covering various domains [17]. In [18], it has been shown that we need a very large training set (∼125,000 hours of semi-supervised speech data) to achieve high speech recognition accuracy for difficult tasks like video captioning.…”

Power-Law Nonlinearity with Maximally Uniform Distribution Criterion for Improved Neural Network Training in Automatic Speech Recognition

“…In the era of deep neural networks, it has been frequently observed that the amount and coverage of the training data seem to be one of the most important factors to obtain better speech recognition accuracy [12,13]. However, it is very difficult to gather sufficient amount of transcribed data from various domains.…”