# k2-fsa/sherpa-ncnn

Real-time speech recognition and voice activity detection (VAD) using next-gen Kaldi with ncnn without Internet connection. Support iOS, Android, Linux, macOS, Windows, Raspberry Pi, VisionFive2, LicheePi4A etc.

Repository: https://github.com/k2-fsa/sherpa-ncnn
Canonical: https://ross.abutalabs.com/products/sherpa-ncnn
Homepage: https://k2-fsa.github.io/sherpa/ncnn/index.html
Language: C++
License: Apache-2.0
License Family: permissive
Topics: python, speech-recognition, cpp, asr, c, csharp, go, kotlin, vad, voice-activity-detection
Last push: 2025-10-20T02:16:19+00:00

## Health v2 (maintenance only)
Score: 58/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 47, release rhythm 48, longevity 100
- inputs: {"age_days": 1459, "days_push": 318, "days_rel": 351, "gap_med": 7, "n_releases_24m": 6}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1779, forks 218 (observed 2026-08-28T04:05:35.311053+00:00)

## What it is
A C++ library for real-time offline speech recognition, text-to-speech, and voice activity detection built on the ncnn inference framework with next-gen Kaldi models. It runs entirely on-device without Internet access and provides APIs for C++, C, Python, Go, C#, Kotlin, Swift, and JavaScript across mobile, desktop, and embedded platforms.

## Use cases
- run offline speech-to-text on android or ios without internet
- real-time microphone transcription on raspberry pi
- add voice activity detection to an embedded device
- on-device text to speech with vits models
- build a wasm browser-based speech recognizer
- streaming asr on risc-v boards like visionfive2

## When to choose
- you need fully offline, low-latency speech recognition on mobile or embedded hardware
- you cannot depend on PyTorch or cloud services and want lightweight ncnn inference
- you need bindings across many languages and platforms including ARM and RISC-V

## When to avoid
- you need GPU-accelerated server-scale ASR with the latest large models
- you want PyTorch-based tooling or training pipelines (use icefall/sherpa-onnx instead)
- you need extensive prebuilt model variety beyond the provided ncnn-exported models

## Facets
- artifact type: library
- maturity: active
- function: speech-recognition, tts, audio-processing, sdk
- domain: speech-processing, embedded-systems, cross-platform
- platform: windows, cpp, python, wasm, embedded, cross-platform
- tags: offline-asr, ncnn, voice-activity-detection, streaming-recognition, raspberry-pi, risc-v, on-device-inference, kaldi, natural-language-processing, android, ios, linux, macos

## Member repositories
- k2-fsa/sherpa-ncnn (main) score 58

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:05:35.311053+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 2026-08-30T03:24:39.491311+00:00, confidence not recorded.
  - readme: https://github.com/k2-fsa/sherpa-ncnn (fetched 2026-08-28T04:05:35.311053+00:00, sha 4308318447fb)
  - homepage: https://k2-fsa.github.io/sherpa/ncnn/index.html (fetched 2026-08-29T11:03:11.750776+00:00, sha a54267d151e2)
- Data as of 2026-08-30T08:39:29.467469+00:00.