CodecFlow: Efficient Bandwidth Extension via Conditional Flow Matching in Neural Codec Latent Space

CodecFlow is a new neural codec-based framework for speech bandwidth extension that efficiently reconstructs high-quality audio in compact latent space. The system uses conditional flow matching and residual vector quantization to improve speech clarity by restoring high-frequency content from low-bandwidth audio.

• →CodecFlow operates in neural codec latent space rather than spectrogram or waveform domains for improved computational efficiency.
• →The framework uses voicing-aware conditional flow converter and structure-constrained residual vector quantizer for better latent alignment.
• →System demonstrates strong spectral fidelity on 8 kHz to 16 kHz and 44.1 kHz speech bandwidth extension tasks.
• →Neural audio codecs provide more compact representations while preserving acoustic detail compared to traditional methods.
• →End-to-end optimization approach achieves enhanced perceptual quality for speech reconstruction applications.