vlac: A lossless audio codec for voice, written in Rust

Description and use

vlac is a lossless codec for 16 kHz mono voice audio, written in Rust, using an Order‑4 LPC predictor and Zstandard with trained dictionaries. It guarantees 100 % bit‑perfect output, compiles to WebAssembly, and supports domain‑specific dictionary training, producing files 8 % larger than FLAC on general voice audio.

Problem

Need a lossless codec that preserves 16kHz mono voice audio with domain‑specific compression and WebAssembly support.

Outcome

vlac achieves 8 % larger files than FLAC on general voice audio but guarantees bit‑perfect output and supports domain‑specific dictionary training and WebAssembly compilation.

Workflow

• Predict next sample using previous three samples (Order‑4 LPC).
• ZigZag encode the residuals.
• Split residuals into High and Low bytes for efficient Zstandard compression of zero‑padded High bytes.
• Compress the split residuals with Zstandard using trained dictionaries.

Evidence

I built vlac, a lossless audio codec designed specifically for 16kHz mono voice audio.

Source

On a 10.75-second voice memo (337 KB WAV): Original WAV: 337 KB, FLAC (ffmpeg default): 151 KB, vlac: 163 KB

Source

vlac currently produces files 8% larger than FLAC on general voice audio. However, it is 100% bit-perfect (verified via MD5 hash on the raw PCM data) and compiles to WebAssembly.

Source