New Speech Codec Cuts Data Use by 90% for Clear Calls
This could mean clearer calls and lower phone bills for everyone.
Researchers propose ESDCodec, a speech codec designed for high-fidelity speech reconstruction at low bitrates. It combines an enhanced semantic quantizer with a conditioned decoder network: a randomly initialized and frozen codebook plus a lightweight projector encode semantic details in a linear space and improve codebook utilization, while a condition network injects prior subband knowledge into the upsampling decoder and predicts subband signals to provide fine-grained guidance for waveform reconstruction. Experiments show ESDCodec achieves superior reconstruction performance at an ultra-low bitrate of 0.85 kbps and outperforms existing codec models on LLM-based speech generation tasks.
- ESDCodec uses 90% less data than normal phone calls, which could save you money on your phone bill.
- It could make calls clearer in areas with poor internet, like rural areas or during network congestion.
- The technology is still in research, so it may be a few years before it's in your devices.
Why It Matters
Clearer calls with less data could lower bills and improve connectivity for millions.