# google/lyra

A Very Low-Bitrate Codec for Speech Compression

Repository: https://github.com/google/lyra
Canonical: https://ross.abutalabs.com/products/lyra
Language: C++
License: Apache-2.0
License Family: permissive
Last push: 2024-08-20T04:55:24+00:00

## Health v2 (maintenance only)
Score: 23/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 0, release rhythm 8, longevity 100
- inputs: {"age_days": 1993, "days_push": 743, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 3973, forks 367 (observed 2026-08-28T04:08:30.899235+00:00)

## What it is
Lyra is a very low-bitrate speech codec from Google that combines traditional codec techniques with generative machine learning models to compress voice at 3.2-9.2kbps. It runs in real time on-device, including low-end Android phones, using a SoundStream-based generative model to reconstruct high-quality speech.

## Use cases
- compress voice calls for very slow networks
- stream speech at ultra-low bitrates
- run real-time speech codec on mobile devices
- encode voice audio at 3kbps with ML
- build low-bandwidth voice communication apps

## When to choose
- you need speech compression far below standard codec bitrates
- you need real-time on-device decoding on mobile hardware
- voice quality at low bitrate matters more than general audio support

## When to avoid
- you need general audio or music compression, not just speech
- you need a standardized, interoperable codec for third-party clients
- you can't integrate Bazel-based C++ builds or ML model dependencies

## Facets
- artifact type: library
- maturity: maintenance
- function: audio-processing, machine-learning, speech-recognition, compression
- domain: machine-learning, speech-processing
- platform: windows, cpp
- tags: speech-codec, low-bitrate, generative-model, soundstream, voice-communication, audio, android, linux, macos

## Member repositories
- google/lyra (main) score 23

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:08:30.899235+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-29T18:24:20.622886+00:00, confidence not recorded.
  - readme: https://github.com/google/lyra (fetched 2026-08-28T04:08:30.899235+00:00, sha 6553e47f43e9)
- Data as of 2026-08-30T08:39:29.467469+00:00.