julius-speech/julius

Health v2 · maintenance only

Adoption not part of the score

1933 stars · 304 forks observed · 2026-08-28

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

Julius is an open-source large vocabulary continuous speech recognition (LVCSR) decoder written in C, based on word N-gram language models and context-dependent HMM acoustic models with a 2-pass tree-trellis search. It supports real-time decoding on devices from microcontrollers to cloud servers, DNN-based decoding, and multi-instance recognition, with model compatibility with HTK and SRILM formats.

Use cases

• transcribe japanese speech to text offline
• run speech recognition on embedded low-memory devices
• decode large vocabulary dictation in real time
• recognize speech using custom grammar or word lists
• experiment with HMM and DNN acoustic models for ASR
• build a voice command recognizer for a robot

When to choose

• you need a small-footprint, self-contained ASR engine in C
• you have HTK/ARPA format acoustic and language models
• you target Linux, Windows, Mac, or Android including embedded devices
• you need grammar-based or isolated-word recognition alongside dictation

When to avoid

• you want a modern pretrained end-to-end model like Whisper out of the box
• you need cloud-scale accuracy on many languages without training your own models
• you prefer Python-first tooling and ecosystems

Facets

library · maturity maintenance

speech-recognition audio-processing nlp speech-processing machine-learning windows cross-platform cli lvcsr hmm dnn-decoding japanese asr c natural-language-processing linux macos android

Member repositories

For agents

markdown · JSON · MCP: product_card(name="julius-speech/julius")