On the Effect of Purely Synthetic Training Data for Different Automatic Speech Recognition Architectures

TL;DR

This paper investigates the effectiveness of synthetic speech data generated by TTS systems for training various ASR architectures, revealing model sensitivities and the impact of data and model variations.

Contribution

It provides a comprehensive evaluation of synthetic data utility for ASR training across multiple architectures and ablation studies on factors affecting performance.

Findings

Synthetic data can effectively train ASR models with varying sensitivity.

Model size and speaker embedding influence synthetic vs. real data gap.

TTS models generalize well despite overfitting in training scores.

Abstract

In this work we evaluate the utility of synthetic data for training automatic speech recognition (ASR). We use the ASR training data to train a text-to-speech (TTS) system similar to FastSpeech-2. With this TTS we reproduce the original training data, training ASR systems solely on synthetic data. For ASR, we use three different architectures, attention-based encoder-decoder, hybrid deep neural network hidden Markov model and a Gaussian mixture hidden Markov model, showing the different sensitivity of the models to synthetic data generation. In order to extend previous work, we present a number of ablation studies on the effectiveness of synthetic vs. real training data for ASR. In particular we focus on how the gap between training on synthetic and real data changes by varying the speaker embedding or by scaling the model size. For the latter we show that the TTS models generalize well, even when training scores indicate overfitting.