Continuous Autoregressive Models with Noise Augmentation Avoid Error Accumulation
This addresses a specific bottleneck in generating continuous embeddings for real-time and interactive applications, representing an incremental improvement.
The paper tackles error accumulation in Continuous Autoregressive Models (CAMs) during sequence generation by injecting random noise into input embeddings during training and inference, resulting in CAM substantially outperforming existing autoregressive and non-autoregressive approaches in musical audio generation while preserving quality over extended sequences.
Autoregressive models are typically applied to sequences of discrete tokens, but recent research indicates that generating sequences of continuous embeddings in an autoregressive manner is also feasible. However, such Continuous Autoregressive Models (CAMs) can suffer from a decline in generation quality over extended sequences due to error accumulation during inference. We introduce a novel method to address this issue by injecting random noise into the input embeddings during training. This procedure makes the model robust against varying error levels at inference. We further reduce error accumulation through an inference procedure that introduces low-level noise. Experiments on musical audio generation show that CAM substantially outperforms existing autoregressive and non-autoregressive approaches while preserving audio quality over extended sequences. This work paves the way for generating continuous embeddings in a purely autoregressive setting, opening new possibilities for real-time and interactive generative applications.