xiph/LPCNet

Health v2 · maintenance only

Flags: no_releases

Adoption not part of the score

1221 stars · 306 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

LPCNet is a low-complexity C implementation of the WaveRNN-based LPCNet neural vocoder for efficient speech synthesis and compression. It applies linear prediction to WaveRNN to achieve real-time neural speech synthesis on modest hardware, and has been used for low-bitrate speech coding and packet loss concealment in Opus.

Use cases

• synthesize speech from acoustic features on CPU
• build a low-bitrate neural speech codec
• implement packet loss concealment for VoIP
• research low-complexity neural vocoders
• run neural speech synthesis in real time without a GPU

When to choose

• you need a proven, low-CPU neural vocoder in C
• you are researching LPC-based neural speech synthesis or coding
• you want to study the techniques later folded into Opus (PLC, deep redundancy)

When to avoid

• you want actively maintained software - the README recommends switching to FARGAN
• you need the best quality-to-complexity ratio, since FARGAN outperforms it at 1/5 the cost
• you want a high-level Python training framework out of the box

Facets

library · maturity abandoned

speech-recognition audio-processing machine-learning deep-learning speech-processing machine-learning c windows cross-platform speech-synthesis vocoder wavernn lpc neural-vocoder low-complexity packet-loss-concealment speech-coding audio linux macos

Member repositories

For agents

markdown · JSON · MCP: product_card(name="xiph/LPCNet")