LDCodec: A high quality neural audio codec with low-complexity decoder
Abstract
Neural audio coding has been shown to outperform classical audio coding at extremely low bitrates. However, the practical application of neural audio codecs is still limited by their elevated complexity. To address this challenge, we have developed a high-quality neural audio codec with a low-complexity decoder, named LDCodec (Low-complexity Decoder Neural Audio Codec), specifically designed for on-demand streaming media clients, such as smartphones. Specifically, we introduced a novel residual unit combined with Long-term and Short-term Residual Vector Quantization (LSRVQ), subband-fullband frequency discriminators, and perceptual loss functions. This combination results in high-quality audio reconstruction with lower complexity. Both our subjective and objective tests demonstrated that our proposed LDCodec at 6kbps outperforms Opus at 12kbps.
The following analysis is generated by AI and may contain errors. It is separate from the authors’ abstract and is not peer review.
Research summary
The main contribution of this paper is the development of LDCodec, a neural audio codec that achieves high-quality audio reconstruction at low complexity, outperforming existing codecs at lower bitrates. This work represents a meaningful advancement in the field of audio processing, particularly for applications requiring efficient audio transmission.
Methodology
The methodology presented in LDCodec is innovative, particularly the introduction of a novel residual unit and the Long-term and Short-term Residual Vector Quantization (LSRVQ). The combination of these elements, along with subband-fullband frequency discriminators and perceptual loss functions, represents a thoughtful approach to reducing complexity while maintaining audio quality. However, the paper lacks detailed descriptions of the implementation of these components, which would enhance understanding and reproducibility.
Results & evaluation
The experiments conducted are robust, with both subjective and objective evaluations demonstrating the codec's performance. The comparison with Opus at different bitrates provides a clear benchmark, and achieving better performance at a lower bitrate is a significant contribution. However, the paper could benefit from a more extensive dataset and additional metrics to further validate the results.
Reproducibility
The paper does not provide sufficient implementation details or code availability, which raises concerns about reproducibility. Clearer guidelines on how to replicate the experiments and access the model would strengthen the paper's impact and utility for other researchers.
Limitations
One limitation is the lack of a comprehensive evaluation across diverse audio types, which may affect the generalizability of the results. Additionally, while the complexity reduction is highlighted, the trade-offs in terms of audio fidelity and potential artifacts in the reconstructed audio are not thoroughly discussed.
Broader impact
The proposed LDCodec has significant implications for on-demand streaming applications, particularly in mobile environments where bandwidth and computational resources are limited. If adopted, it could enhance user experiences in audio streaming services, making high-quality audio more accessible. The main contribution of this paper is the development of LDCodec, a neural audio codec that achieves high-quality audio reconstruction at low complexity, outperforming existing codecs at lower bitrates. This work represents a meaningful advancement in the field of audio processing, particularly for applications requiring efficient audio transmission.
Author of this paper? Send a correction.