Saved in:
| Main Authors: | , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2026
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2605.19541 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866913146117029888 |
|---|---|
| author | Wang, Junyi Zhang, Chi Qian, Jing Luo, Haifeng Wang, Hao Jin, Zengrui Zhang, Chao |
| author_facet | Wang, Junyi Zhang, Chi Qian, Jing Luo, Haifeng Wang, Hao Jin, Zengrui Zhang, Chao |
| contents | In bandwidth-constrained communication such as satellite and underwater channels, speech must often be transmitted at ultra-low bitrates where intelligibility is the primary objective. At such extreme compression levels, codecs trained with acoustic reconstruction losses tend to allocate bits to perceptual detail, leading to substantial degradation in word error rate (WER). This paper proposes ClariCodec, a neural speech codec operating at 300 bit per second (bps) that reformulates quantisation as a stochastic policy, enabling reinforcement learning (RL)-based optimisation of intelligibility. Specifically, the encoder is fine-tuned using WER-driven rewards while the acoustic reconstruction pipeline remains frozen. Even without RL, ClariCodec achieves 4.64% WER on the LibriSpeech test-clean set at 300 bps, already competitive with codecs operating at higher bitrates. Further RL fine-tuning reduces WER to 3.55% on test-clean and 10.4% on test-other, corresponding to a 23% relative reduction while preserving perceptual quality. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2605_19541 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Optimising Neural Speech Codecs for 300bps Communication using Reinforcement Learning Wang, Junyi Zhang, Chi Qian, Jing Luo, Haifeng Wang, Hao Jin, Zengrui Zhang, Chao Sound In bandwidth-constrained communication such as satellite and underwater channels, speech must often be transmitted at ultra-low bitrates where intelligibility is the primary objective. At such extreme compression levels, codecs trained with acoustic reconstruction losses tend to allocate bits to perceptual detail, leading to substantial degradation in word error rate (WER). This paper proposes ClariCodec, a neural speech codec operating at 300 bit per second (bps) that reformulates quantisation as a stochastic policy, enabling reinforcement learning (RL)-based optimisation of intelligibility. Specifically, the encoder is fine-tuned using WER-driven rewards while the acoustic reconstruction pipeline remains frozen. Even without RL, ClariCodec achieves 4.64% WER on the LibriSpeech test-clean set at 300 bps, already competitive with codecs operating at higher bitrates. Further RL fine-tuning reduces WER to 3.55% on test-clean and 10.4% on test-other, corresponding to a 23% relative reduction while preserving perceptual quality. |
| title | Optimising Neural Speech Codecs for 300bps Communication using Reinforcement Learning |
| topic | Sound |
| url | https://arxiv.org/abs/2605.19541 |