Repository navigation
v0.7.0
v0.7.0 - 首段 AI 输出提速与分句链路完善
中文
亮点
首段 AI 输出更快
- 首段完成即触发 AI，并持续写入文本框，显著降低停录后的等待时间。
- 语音转录后的 AI 触发策略优化为更小的分句阈值（按 2 句触发）。
分句与分组更稳定
- 引入轻量分句器，提升无标点文本的分句稳定性，避免长句卡住首段输出。
基准与可观测性
- 新增首段输出基准链路，支持复用 ASR 模型的跑批模式，减少基准耗时。
升级步骤
从 v0.6.4 升级
- 下载
SonicInput-v0.7.0-win64.exe。 - 关闭旧版本并替换可执行文件。
- 如需离线包，下载对应
-offline.zip。
支持
- 配置：
%AppData%/SonicInput/config.json - 日志：
%AppData%/SonicInput/logs/app.log - Issues：https://github.com/Oxidane-bot/SonicInput/issues
English
Highlights
Faster first AI output
- Trigger AI as soon as the first chunk is ready, then keep streaming into the text box to reduce post-stop latency.
- Smaller sentence grouping threshold (2 sentences) for earlier AI triggers.
More stable sentence splitting
- Lightweight sentence splitter improves segmentation for punctuation-free text, preventing long-blocking chunks.
Bench & observability
- Added first-output benchmark path with ASR model reuse to reduce batch runtime.
Upgrade Steps
From v0.6.4
- Download
SonicInput-v0.7.0-win64.exe. - Close the old version and replace the executable.
- For offline usage, download the matching
-offline.zip.
Support
- Config:
%AppData%/SonicInput/config.json - Logs:
%AppData%/SonicInput/logs/app.log - Issues: https://github.com/Oxidane-bot/SonicInput/issues