Low Frame-rate Speech Codec

Abstract:

Large language models (LLMs) have significantly advanced audio processing through audio codecs that convert audio into discrete tokens, enabling the application of language modeling techniques to audio data. However, audio codecs often operate at high frame rates, resulting in slow training and inference, especially for autoregressive models. To address this challenge, we present the Low Frame-rate Speech Codec (LFSC): a neural audio codec that leverages finite scalar quantization and adversarial training with large speech language models to achieve high-quality audio compression with a 1.89 kbps bitrate and 21.5 frames per second. We demonstrate that our novel codec can make the inference of LLM-based text-to-speech models around three times faster while improving intelligibility and producing quality comparable to previous models.

Codec Reconstruction

Note that for the audio demo all audios was resampled to 22.05kHz to fair comparison.

MLS test set

English samples

Portuguese samples

French samples

Dutch samples

German samples

Italian samples

Polish samples

Portuguese samples

Spanish samples

DAPS test set

ZS-TTS Audio Samples

Note that for the audio demo all audios was resampled to 22.05kHz to fair comparison.

Citation

        @article{casanova2024low,
          title={Low Frame-rate Speech Codec: a Codec Designed for Fast High-quality Speech LLM Training and Inference},
          author={Casanova, Edresson and Langman, Ryan and Neekhara, Paarth and Hussain, Shehzeen and Li, Jason and Ghosh, Subhankar and Juki{\'c}, Ante and Lee, Sang-gil},
          journal={arXiv preprint arXiv:2409.12117},
          year={2024}
        }