Release 3.13.1

📝 更新日志 / Changelog (v3.13.1)

中文

新增功能

• Shizuku/Root 增强保活: 添加 Shizuku/Root 增强保活功能，提升后台运行稳定性
• 内置提示词多语言支持: 实现内置提示词的多语言支持，提升不同语言用户的使用体验
• LLM 测试取消: LLM 测试时可以取消以打断测试
• [Pro]新增AI助手功能: 在 Pro 版本中添加 AI 助手功能，通过关键词触发调用
• [Pro]热词注入优化: 启用热词注入 AI 后处理提示词后，自动基于音素匹配进行热词筛选，通过提升热词注入的准确性来优化模型后处理的效果

改进优化

• 悬浮球录音动画: 优化悬浮球录音呼吸动画效果
• LLM 温度设置: 更新默认温度值并简化温度标签显示

---

English

New Features

• Shizuku/Root Enhanced Keep-Alive: Added Shizuku/Root enhanced keep-alive feature for improved background stability
• Built-in Prompts Multilingual Support: Implemented multilingual support for built-in prompts, enhancing user experience for different language speakers
• LLM Test Cancellation: Added LLM test cancellation feature
• [Pro] New AI Assistant Feature: Added AI assistant feature in Pro version, triggered by keywords
• [Pro] Hotword Injection Optimization: When enabling hotword injection AI post-processing

Improvements

• Floating Ball Recording Animation: Optimized floating ball recording breathing animation effect
• LLM Temperature Settings: Updated default temperature value

Release 3.13.0

📝 更新日志 / Changelog (v3.13.0)

中文

新增功能

• 设置搜索功能: 新增设置搜索功能
• 历史记录优化: 在历史记录中新增外部输入来源类别标识，优化处理时间显示
• AI编辑面板优化: 为AI编辑面板添加信息栏和空格键，优化整体布局
• Soniox模型升级: 更新Soniox ASR非流式引擎使用的模型版本至v4

改进优化

• Soniox语言选择器优化: 优化Soniox语言选择器UI
• UI细节优化: 粘贴板面板图标优化，调整数字键盘面板高度和边距
• 架构重构优化: 重构键盘核心、ASR设置、悬浮球交互及设置存储模块，提升代码可维护性
• VAD性能优化: 实现VAD实例池化，提升性能和资源利用率
• RecognitionService增强: 改进RecognitionService服务的安全性和兼容性
• 输入法磁贴重构: 重构输入法选择器磁贴逻辑

问题修复

• 修复悬浮球错误状态动画和视觉效果的清理问题
• 为流式ASR引擎补齐VAD资源释放逻辑
• 修复AI处理失败时的识别记录处理问题

变更说明

• 移除豆包语音双向流式和首字加速的实验性功能

---

English

New Features

• Settings Search: Added settings search functionality
• History Optimization: Added external input source category identifier in history records with optimized time display
• AI Edit Panel Enhancement: Added info bar and space key to AI edit panel with optimized layout
• Soniox Model Upgrade: Updated Soniox ASR non-streaming engine model version to v4

Improvements

• Soniox Language Selector Optimization: Optimized Soniox language selector UI
• UI Detail Optimization: Optimized pasteboard panel icon and adjusted numeric keyboard panel height and margins
• Architecture Refactoring: Refactored keyboard core, ASR settings, floating ball interaction and settings storage modules for better maintainability
• VAD Performance Optimization: Implemented VAD instance pooling to improve performance and resource utilization
• RecognitionService Enhancement: Improved security and compatibility of RecognitionService
• IME Tile Refactoring: Refactored input method selector tile logic

Bug Fixes

• Fixed floating ball error state animation and visual effect cleanup issues
• Added VAD resource release logic for streaming ASR engines
• Fixed recognition record handling when AI processing fails

Changes

• Removed experimental features of Doubao voice bidirectional streaming and first-character acceleration

Release 3.12.1

📝 更新日志 / Changelog (v3.12.1)

中文

新增功能

• 悬浮球图标优化: 优化悬浮球图标和动画效果，常驻模式下贴边半隐时显示箭头把手
• 历史记录统计增强: 增强历史记录统计信息，新增AI后处理耗时显示
• 外部输入法联动增强: 改进外部输入法联动指引，新增支持 Trime 输入法

改进优化

• AI后处理打字机效果: 仅在AI后处理真正生效时使用打字机动画

问题修复

• 增加贴边锚点系统，解决横竖屏切换时悬浮球位置偏移问题
• 修复使用安卓导航栏时选项菜单显示不全的问题
• 修复冷启动时系统 insets 异常导致的布局问题

---

English

New Features

• Floating Ball Icon Optimization: Optimized floating ball icon and animation effects, added arrow handle when floating ball is docked at screen edge in persistent mode
• History Statistics Enhancement: Enhanced history statistics with AI post-processing time display
• External IME Linkage Enhancement: Improved external input method linkage guidance with added support for Trime input method

Improvements

• AI Post-Processing Typewriter Effect: Only use typewriter animation when AI post-processing is actually active

Bug Fixes

• Added edge anchor system to fix position offset issues when switching between portrait and landscape modes
• Fixed options menu display issue when using Android navigation bar
• Fixed layout issues caused by abnormal system insets during cold start

Release 3.12.0

📝 更新日志 / Changelog (v3.12.0)

中文

新增功能

• 并行主备双引擎识别: 新增并行主备双引擎识别功能，提升语音识别准确性和可靠性
• FunASR Nano 独立供应商: 将 FunASR Nano 从 SenseVoice 中剥离，新增为独立的 ASR 供应商
• FunASR Nano 推理优化: 支持 FunASR Nano 完整版本，提升识别精度
• 后处理打字机效果: 为流式 AI 后处理结果添加打字机动画效果，提升交互体验
• 后处理打字机开关: 在后处理设置中新增打字机效果开关选项
• 切换输入法磁贴: 新系统设置快捷磁贴，方便快速切换输入法
• 增加供应商标签: 为语音识别供应商增加标签，帮助用户选择合适的供应商
• [Pro] 一键删除热词: 在热词管理界面添加一键删除所有热词功能
• [Pro] 正则后处理优化: 内置常用正则表达式模板，支持配置备注说明，优化UI

改进优化

• 本地模型加载协调器: 引入本地模型加载协调器，优化模型加载流程，提升加载可靠性
• 优化 ITN 效果: 优化中文数字归一化效果，提升数字识别准确性
• Soniox 流式引擎重构: 修复 Soniox 流式引擎的若干问题
• ASR 选项显示名称: 精简多 ASR 选项显示名称，提升界面简洁度

问题修复

• 尝试修复 Paraformer 句尾丢字的问题
• 避免本地模型首次加载时误触发超时

---

English

New Features

• Parallel Primary-Backup Dual-Engine Recognition: Added parallel primary-backup dual-engine recognition feature to improve speech recognition accuracy and reliability
• FunASR Nano Independent Vendor: Separated FunASR Nano from SenseVoice as an independent ASR vendor
• FunASR Nano Inference Optimization: Added support for FunASR Nano full version to improve recognition accuracy
• Post-Processing Typewriter Effect: Added typewriter animation effect for streaming AI post-processing results to enhance interaction experience
• Post-Processing Typewriter Toggle: Added typewriter effect toggle option in post-processing settings
• Switch IME Tile: Added new system settings quick tile for convenient input method switching
• Vendor Tags: Added tags for speech recognition vendors to help users choose the appropriate vendor
• [Pro] One-Tap Hotword Deletion: Added one-tap delete all hotwords feature in hotword management interface
• [Pro] Regex Post-Processing Optimization: Built-in common regex templates with configurable descriptions and optimized UI

Improvements

• Local Model Loading Coordinator: Introduced local model loading coordinator to optimize model loading process and improve loading reliability
• Optimized ITN Effect: Optimized Chinese number normalization effect to improve number recognition accuracy
• Soniox Streaming Engine Refactor: Fixed several issues with Soniox streaming engine
• ASR Option Display Names: Simplified multi-ASR option display names to improve interface cleanliness

Bug Fixes

• Attempted to fix Paraformer end-of-sentence character loss issue
• Prevented false timeout trigger on first local model load

Release 3.11.2

📝 更新日志 / Changelog (v3.11.2)

中文

新增功能

• 触觉反馈强度调节: 新增触觉反馈强度分级调节功能，并补全各场景的触觉反馈
• 自定义推理参数: 新增自定义思考模式的参数配置功能
• 前台保活服务: 其他设置页新增前台保活服务及电池优化白名单申请入口
• 下载源测速: 下载源选择弹窗增加延迟测速显示，方便选择最快的镜像源
• 自动检查更新: 关于页新增"自动检查更新"开关选项
• 更新内置后处理模型列表: 为部分渠道新增多个内置后处理模型可供选择

改进优化

• 预置渠道拉取模型: 将拉取模型列表功能扩展到预置渠道，不再仅限于自定义渠道
• 下载镜像源配置: 统一下载镜像源配置并简化调用逻辑

问题修复

• 修复本地模型"配置指南"跳转按钮失效的问题
• 修复同步总字符数时可能丢失已统计 ASR 字符的问题

---

English

New Features

• Haptic Feedback Intensity Adjustment: Added haptic feedback intensity level adjustment and completed haptic feedback for various scenarios
• Custom Thinking Parameters: Added custom thinking parameter configuration for AI post-processing models
• Foreground Keep-Alive Service: Added foreground keep-alive service and battery optimization whitelist request entry in Other Settings
• Download Source Speed Test: Added latency speed test display in download source selection dialog for easier fastest mirror selection
• Auto Update Check: Added "Auto Check for Updates" toggle option in About page
• New Built-in Post-Processing Models: Added multiple built-in post-processing models to choose from

Improvements

• Preset Channel Model Fetch: Extended model list fetching to preset channels, no longer limited to custom channels only
• Download Mirror Configuration: Unified download mirror source configuration and simplified calling logic

Bug Fixes

• Fixed "Configuration Guide" jump button not working for local models
• Fixed potential loss of already counted ASR characters when syncing total character count

Release 3.11.1

📝 更新日志 / Changelog (v3.11.1)

中文

新增功能

• 火山引擎新模型支持: 新增火山引擎 DeepSeek-V3-2-251201 模型支持
• [Pro] 应用专属 Prompt 设置: 在 AI 后处理设置中添加应用专属 Prompt 配置功能，借助无障碍能力允许为不同应用设置定制化的后处理提示词

改进优化

• AI 后处理流式输出: 为 AI 后处理增加流式结果输出功能,用户可在等待最终结果时先看到逐步生成的文本片段,提升交互体验
• 设置选项底部弹窗: 新增设置选项底部弹窗组件并替换原有对话框,提升交互顺畅性
• 字数统计性能优化: 使用区间判断优化 CJK 字符检测性能,并修复部分极端文本下的统计错误问题

变更说明

• 移除 Zipformer 引擎: 为精简本地 ASR 方案,下线 Zipformer 流式引擎。现有用户将自动迁移至 Paraformer;需在 ASR 设置中重新确认或切换模型。已在 AsrVendor 枚举中保留 "zipformer" id 到 Paraformer 的映射,保证旧配置平滑迁移,并新增一次性清理逻辑,首次启动时移除已下载的 zipformer 模型目录

---

English

New Features

• Volcengine New Model Support: Added Volcengine DeepSeek-V3-2-251201 model support
• [Pro] App-Specific Prompt Settings: Added app-specific prompt configuration feature in AI post-processing settings, allowing customized post-processing prompts for different apps using accessibility capabilities

Improvements

• AI Post-Processing Streaming Output: Added streaming output functionality for AI post-processing, allowing users to see progressively generated text fragments while waiting for the final result, improving interactive experience
• Settings Option Bottom Sheet: Added new settings option bottom sheet component to replace original dialogs, enhancing interaction smoothness
• Word Count Performance Optimization: Optimized CJK character detection performance using range checks and fixed counting errors in some extreme text cases

Changes

• Removed Zipformer Engine: Deprecated Zipformer streaming engine to streamline local ASR solutions. Existing users will be automatically migrated to Paraformer; please reconfirm or switch models in ASR settings. The "zipformer" ID mapping to Paraformer is preserved in the AsrVendor enum to ensure smooth legacy configuration migration, with one-time cleanup logic added to remove downloaded zipformer model directories on first launch

Release 3.11.0

📝 更新日志 / Changelog (v3.11.0)

中文

新增功能

• 应用图标更新: 更新应用图标设计并适配 Monet 动态色系统
• 悬浮球拖动优化: 支持直接拖动移动悬浮球选项，无需长按进入拖动模式
• 输入法切换增强: 允许手动指定切回的目标输入法，提升多输入法切换体验
• 自定义 LLM 增强: 自定义供应商支持无需模型名配置，简化连接验证流程
• LLM 模型列表: 为自定义后处理渠道新增「拉取模型列表」功能

改进优化

• 图标资源重构: 重构应用图标资源结构并精简颜色定义
• 文案优化: 明确 Play Store 兑换流程文案

问题修复

• 修复伪流式缓冲区处理流程问题

---

English

New Features

• App Icon Update: Updated app icon design with Monet dynamic color system adaptation
• Floating Ball Drag Enhancement: Added direct drag-to-move support for floating ball, no need to long-press to enter drag mode
• IME Switching Enhancement: Added manual target IME selection when switching back, improving multi-IME switching experience
• Custom LLM Enhancement: Custom vendors now support configuration without model names, simplifying connection validation
• LLM Model List: Added "Fetch Model List" feature for custom post-processing channels

Improvements

• Icon Resource Refactoring: Refactored app icon resource structure and streamlined color definitions
• Wording Optimization: Clarified Play Store redemption process wording

Bug Fixes

• Fixed pseudo-streaming buffer processing flow issue

Release 3.10.1

📝 更新日志 / Changelog (v3.10.1)

中文

新增功能

• VAD 引擎升级: 将语音活动检测引擎由 Silero VAD 替换为 Ten VAD，提升静音检测准确性
• Gemini 自定义端点: 支持自定义 Gemini API endpoint，方便使用代理或私有部署
• 非流式降噪功能: 为非流式识别引擎添加降噪功能开关，提升嘈杂环境下的识别准确性

改进优化

• 伪流式识别架构: 抽离 SenseVoice 与 TeleSpeech 伪流式公共逻辑至独立 Delegate，提升代码可维护性
• 伪流式预览策略: 改用"定时分片+VAD过滤"策略进行定时伪流式预览，优化预览效果

问题修复

• 修复静音自动停止在本地模型伪流式模式下失效的问题
• 修复火山引擎识别模型版本选择问题
• 长按录音模式下禁止静音自动停录，避免误操作

---

English

New Features

• VAD Engine Upgrade: Replaced Silero VAD with Ten VAD for improved voice activity detection accuracy
• Gemini Custom Endpoint: Added support for custom Gemini API endpoint, facilitating proxy or private deployment usage
• Non-Streaming Noise Reduction: Added noise reduction toggle for non-streaming recognition engines to improve accuracy in noisy environments

Improvements

• Pseudo-Streaming Architecture: Refactored SenseVoice and TeleSpeech pseudo-streaming common logic into an independent Delegate for improved maintainability
• Pseudo-Streaming Preview Strategy: Adopted "timed segmentation + VAD filtering" strategy for pseudo-streaming preview to optimize preview effects

Bug Fixes

• Fixed silence-based auto-stop not working in local model pseudo-streaming mode
• Fixed Volcengine recognition model version selection issue
• Disabled silence auto-stop in long-press recording mode to prevent accidental stops

Release 3.10.0

📝 更新日志 / Changelog (v3.10.0)

中文

新增功能

• 本地 ASR 引擎标点功能: 在多个本地 ASR 引擎（FunASR Nano、TeleSpeech、Paraformer、Zipformer）中增加标点功能，提升离线识别的可用性
• FunASR Nano 模型支持: 添加 FunASR Nano 模型支持
• 通用标点模型管理: 新增通用标点模型管理系统，支持标点模型的统一管理和版本控制
• Soniox 语言严格限制模式: 为 Soniox ASR 引擎添加语言严格限制模式选项，提升多语言环境下的识别准确性
• Dashscope Fun-ASR 语义断句: 为 DashScope Fun-ASR 添加语义断句选项，替代 VAD 断句

改进优化

• sherpa-onnx 升级: 升级 sherpa-onnx 至 1.12.20 版本
• SenseVoice 模型版本调用: 优化 SenseVoice Small 模型的版本调用逻辑，确保模型加载的准确性

---

English

New Features

• Local ASR Engine Punctuation Support: Added punctuation functionality to multiple local ASR engines (FunASR Nano, TeleSpeech, Paraformer, Zipformer) to improve offline recognition readability
• FunASR Nano Model Support: Added FunASR Nano model support
• Universal Punctuation Model Management: Added universal punctuation model management system with unified management and version control for punctuation models
• Soniox Language Strict Restriction Mode: Added language strict restriction mode option for Soniox ASR engine to improve recognition accuracy in multilingual environments
• Dashscope Fun-ASR Semantic Segmentation: Added semantic segmentation option for DashScope Fun-ASR

Improvements

• sherpa-onnx Upgrade: Upgraded sherpa-onnx to version 1.12.20
• SenseVoice Model Version Calling: Optimized SenseVoice Small model version calling logic to ensure accurate model loading

Release 3.9.4

📝 更新日志 / Changelog (v3.9.4)

中文

新增功能

• Fun-ASR-Realtime 模型: 为 DashScope 新增 Fun-ASR-Realtime 模型支持，提升实时语音识别性能
• 优化初始判停延迟: 提高初始判停延迟，提升语音识别的用户体验
• 官网与文档按钮: 在关于页面新增官网和文档按钮，方便用户获取帮助和了解产品
• [Pro] Fun-ASR-Realtime 热词适配: 为 DashScope Fun-ASR-Realtime 模型添加热词适配

改进优化

• DashScope 模型选择器: 重构 DashScope ASR 模型选择器，优化模型管理和切换体验
• 键盘稳定性: 确保软键盘弹出与启用状态判断的稳定性，提升输入法可靠性

问题修复

• 本地流式模型稳定性: 修复本地流式模型的识别稳定性问题，提升识别准确性和连续性
• 更新安装页面: 修复更新安装完成后回到应用又再次弹出安装页面的问题

---

English

New Features

• Fun-ASR-Realtime Model: Added Fun-ASR-Realtime model support for DashScope to improve real-time speech recognition performance
• Initial VAD Delay: Improved initial Voice Activity Detection (VAD) delay to enhance speech recognition user experience
• Website and Documentation Buttons: Added official website and documentation buttons to the about page for easy user access to help and product information
• [Pro] Fun-ASR-Realtime Hotword Adaptation: Added hotword adaptation for DashScope Fun-ASR-Realtime model

Improvements

• DashScope Model Selector: Refactored DashScope ASR model selector architecture for improved model management and switching experience
• Keyboard Stability: Ensured stability of soft keyboard popup and enabled state detection for enhanced input method reliability

Bug Fixes

• Local Streaming Model Stability: Fixed stability issues with local streaming model recognition, improving recognition accuracy and continuity
• Update Installation Page: Fixed issue where the update installation page would pop up again after returning to the app following completion