niyongsheng/AudioScript

AudioScript is worth a look if you regularly watch foreign-language video or podcasts on a Mac and want a floating bilingual subtitle window without sending audio to a cloud service. The README is short; the practical trade-off is that its language pair and platform are narrower than its description might suggest.

The problem

Say you are watching a conference talk, a foreign podcast, or a video with no subtitles. Ordinarily you either find a transcript afterward or route the audio through a separate service. AudioScript instead captures your system audio output (or the microphone, or a mix of both) and turns it into on-screen captions as the audio plays. You configure the source and see original text plus a Chinese translation in a floating window, with adjustable font size, opacity, and line count.

What's different

The mechanism is the part that matters: streaming speech recognition through sherpa-onnx, paired with CTranslate2 running an opus-mt translation model, both locally. In practice, that means no audio has to leave the machine for recognition or translation, which is exactly what makes it usable for meetings or material you would not upload elsewhere. The README states the translation direction is English to Chinese and back, so this is not a general-purpose multilingual captioner. The floating overlay is built on a Tauri panel layer, and the app includes tray behavior, close-to-tray, single-instance, and launch-at-login — details that matter for something meant to sit on screen during a call. There is an Aptabase usage-tracking integration, which is disclosed and can be turned off in settings; that is a reasonable default only if you find the toggle.

Who should try it

Good fit: a Mac user on macOS 14.2 or later who watches or listens to English material and wants Chinese subtitles live, especially where privacy or offline use matters. Poor fit: someone needing transcription in languages other than English and Chinese, or someone who wants a post-recording transcript/editor workflow rather than a live overlay — the README describes real-time captioning, not a transcript editor. The licensing also matters if you plan to redistribute or embed it: the project is MIT, but the underlying sherpa-onnx, CTranslate2, and opus-mt components carry their own terms.

Requirements and limitations

The README badges list macOS 14.2+ only, and there is no documented Windows or Linux build path, so the "cross-platform" topic tag should not be read as a current claim. Building from source requires pnpm, a Tauri 2 toolchain, and downloading about 58 MB of ASR models plus the translation models, so this is not a drop-in developer dependency. Running it also depends on the local model files being present; if the download step fails, there is no documented cloud fallback. The README does not document accuracy, latency, or how well translation handles domain-specific vocabulary, so evaluation will have to come from your own material. Ongoing upkeep means watching the model-download step and the Tauri side over time; the README does not describe an update mechanism.

HotRepo's take

HotRepo has only one dated observation: on 2026-10-27 the project, then 13 days old, appeared at #25 on the NEW-REPOSITORY board for Rust with 237 total stars. That is a single snapshot on a newcomer board, not a growth series, so persistence cannot yet be judged. Treat it as an early-stage app with a clear use case. Before relying on it, verify that its English-to-Chinese model handles your actual audio well and confirm the macOS requirement and any non-English needs against the README.