Versions:
Chough 0.2.5, published by hyperpuncher, is a lightweight command-line automatic-speech-recognition utility built on the memory-efficient sherpa-onnx framework and driven by the high-speed Parakeet TDT 0.6b V3 Fast acoustic model. Designed for developers, researchers, and system administrators who need to transcribe spoken audio without a graphical interface, the program ingests common audio formats, processes them in configurable chunks, and streams accurate text to stdout or a nominated file. Its chunked architecture keeps RAM usage low even on hour-long recordings, while the underlying ONNX runtime leverages CPU or optional GPU acceleration to deliver real-time factors well below 1× on modern hardware. Typical use cases include batch-subtitling of lecture archives, generating meeting minutes from VoIP exports, adding text metadata to podcast repositories, and feeding live captioning pipelines. Because all parameters—model path, chunk length, language tokens, and beam size—are exposed as CLI flags, Chough slots easily into shell scripts, CI workflows, or Docker containers that must remain slim and dependency-free. The project has iterated twice since inception, with version 0.2.5 refining punctuation insertion, reducing initialization latency, and adding a 16-bit integer audio path for embedded targets. The software is available for free on get.nero.com, with downloads provided via trusted Windows package sources (e.g. winget), always delivering the latest version, and supporting batch installation of multiple applications.
Tags: