# kyutai-labs/hibiki

Hibiki is a model for streaming speech translation (also known as simultaneous translation). Unlike offline translation—where one waits for the end of the source utterance to start translating--- Hibiki adapts its flow to accumulate just enough context to produce a correct translation in real-time, chunk by chunk.

Repository: https://github.com/kyutai-labs/hibiki
Canonical: https://ross.abutalabs.com/products/hibiki
Language: Rust
License: Apache-2.0
License Family: permissive
Last push: 2025-04-15T04:29:05+00:00

## Health v2 (maintenance only)
Score: 28/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 16, release rhythm 35, longevity 41
- inputs: {"age_days": 575, "days_push": 505, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1509, forks 119 (observed 2026-08-28T04:04:55.702627+00:00)

## What it is
Hibiki is a decoder-only model for streaming (simultaneous) speech-to-speech translation, built on the multistream Moshi architecture. It processes source speech in real time and generates target-language speech with optional voice transfer, along with timestamped text translation.

## Use cases
- translate live speech in real time while the speaker is still talking
- simultaneous interpretation for meetings or broadcasts
- generate translated speech preserving the original speaker's voice
- produce timestamped text transcripts of a translation as audio streams
- run a speech translation model locally on GPU

## When to choose
- you need low-latency, chunk-by-chunk translation rather than waiting for the utterance to end
- you want both audio and text output from a single model
- voice transfer to keep the speaker's voice in the translation matters
- you need a batch-friendly model with simple temperature sampling inference

## When to avoid
- you only need offline translation of complete recordings with maximum quality
- your language pair is not supported by the released models
- you have no GPU and need fast real-time performance
- you need text-only document translation

## Facets
- artifact type: library
- maturity: active
- function: speech-recognition, machine-learning, audio-processing, llm-inference
- domain: speech-processing, machine-learning
- platform: python, rust, cross-platform
- tags: simultaneous-translation, speech-to-speech, streaming-translation, voice-transfer, moshi, decoder-only, natural-language-processing, audio, gpu

## Member repositories
- kyutai-labs/hibiki (main) score 28

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:04:55.702627+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 2026-08-30T04:32:29.849669+00:00, confidence not recorded.
  - readme: https://github.com/kyutai-labs/hibiki (fetched 2026-08-28T04:04:55.702627+00:00, sha 51007e59757a)
  - registry_crates: https://crates.io/api/v1/crates/hibiki (fetched 2026-08-29T11:36:36.325382+00:00, sha 0bca6ca55ceb)
- Data as of 2026-08-30T08:39:29.467469+00:00.