v0.7.5 — 认得出：翻译不再无声死掉，麦克风不再录到静音 Recognition & recovery

中文

• 翻译可能在一整场会议里彻底消失：启动时握手失败一次就把翻译通道锁死，之后再无重试，只有最开头一条提示然后一片空白——同一天、同样设置的另一场却正常出双语。现在翻译通道自动重连，真连不回来时状态栏会写「翻译失败」，不再装作还在工作
• 桌面版选「麦克风」模式录到的全是静音。桌面版的系统识别走的是 CoreAudio 的系统输出捕获，根本没有麦克风通路；选了麦克风只是在输出捕获上贴了个麦克风的标签，音频帧在涨、峰值始终是零、什么都识别不出来。麦克风模式现在走本地模型。之前已经保存了错误配置的安装会在启动时自动修正
• 设置里的引擎卡片只写了引擎没写音频来源：选「系统识别」时来源还显示「麦克风」，底栏声称麦克风在录，实际走的是系统音频捕获
• 本地转录服务的端口（8765/8766）被上一次强退或第二个 app 实例占着时，启动直接崩出一段 Python 报错。现在提示是人话：哪个端口被占了、大概率是什么进程、怎么杀掉
• 设置和首次运行教程把音频来源当识别引擎来列，还在写「本地 Whisper」——实际本地模型也跑 Parakeet。术语已全部校正
• 系统识别引擎接受一组提示词来提高专业术语的转录准确率，以前平均分配给所有启用的词典包，一场生信报告大部分预算花在了金融和日常词汇上，真正的生信术语只分到大约十个名额。现在按当前会议的领域加权。个人生词本以前会独占整个预算，现在和词典包共享。这只提高已有词条被正确转录的概率，不等于新增了词条——湿实验/蛋白/免疫学包还没做
• 会议记录一键导出到 Obsidian 和 Notion
• 详见 https://mianaz.github.io/jargonslayer/

English

• Translation could die for an entire meeting: a single failed startup handshake latched the translation lane off with no retry, one toast at the very start and then nothing — while another session the same day with the same settings produced full bilingual output. The lane now retries on its own, and if recovery genuinely fails the status bar reads 翻译失败 instead of looking like it is still working
• Desktop microphone mode recorded silence. The desktop system recognizer is a CoreAudio system-output tap with no microphone path; selecting microphone started the output tap under a mic label — audio frames climbed, peak stayed at zero, nothing was recognized. Microphone mode now routes to the local model. Installs that were already saved in the broken state self-heal on launch
• The settings engine cards wrote the engine without its paired audio source, so picking 系统识别 while the source read 麦克风 left the bottom bar claiming the mic was live over a system tap
• A stale local transcription service on ports 8765/8766 — a force-quit leftover or a second app instance — crashed startup with a raw Python traceback. The error now names the port, the likely stale process, and the command to kill it
• Settings and the first-run tutorial listed audio sources as if they were recognizers, and still said 本地 Whisper for a local model that also runs Parakeet. Swept
• The system recognizer accepts a limited list of hint terms to improve specialist transcription. That list was split evenly across every enabled dictionary pack, so a bioinformatics talk spent most of its budget on finance and everyday vocabulary and got roughly ten slots of bioinformatics file formats. Hints are now weighted toward the domain the current meeting is actually about. A large personal glossary also used to consume the entire budget before any pack term was reached; glossary and packs now share it. This improves the odds a specialist term is transcribed at all — it does not add new dictionary entries; the wet-lab / protein / immunology pack is still to come
• One-click session export to Obsidian and Notion
• Details: https://mianaz.github.io/jargonslayer/

校验和 / Checksums

JargonSlayer-0.7.5.dmg  sha256: 613237934fa6a612b2156809d3dfb223e4d63c7da4346dd57707ccffda65bf14
jargonslayer-extension-0.7.5.zip  sha256: 101c9b930986e47dcd9fd7c32354474989b8f67a0e28cb89b3f50df6eabc59ad