📝 更新日志 / Changelog (v4.0.1)

中文

新增功能

• 悬浮键盘模式：支持悬浮键盘和对应控制开关
• 宽屏优化：横屏及平板设备优化显示，并支持左右停靠
• 标点位置跳转按钮：键盘新增基于标点位置的前后移动跳转按钮
• fxliang 小企鹅语音联动：支持 fxliang 版小企鹅输入法通过 AIDL 调用语音识别
• 句末标点无限制选项：句末标点与 emoji 去除阈值新增“无限制”选项

改进优化

• Soniox 模型升级：升级 Soniox 非流式识别模型至 v5
• 触觉反馈重构：重构触觉反馈机制并集中管理
• 录音测试页优化：优化录音测试页的延迟颜色表现
• [Pro] AMOLED 独立开关：将 AMOLED 模式重构为独立开关，旧模式配置自动迁移
• [Pro] 自动更新检测优化：优化自动更新检查，关闭后可以不依赖 Play 服务

问题修复

• 键盘底部抬高：优化部分场景下的键盘底部抬高间隙
• ITN 优化：优化 ITN 在部分场景的表现
• [Pro] 热词统计修复：修复热词命中次数统计与颜色分级

English

New Features

• Floating Keyboard Mode: Added floating keyboard support with a control toggle
• Widescreen Optimization: Optimized display for landscape and tablet devices, with left/right docking support
• Punctuation Jump Button: Added forward/backward jump buttons in the keyboard based on punctuation positions
• fxliang Fcitx5 Voice Integration: Support for fxliang's Fcitx5 IME to invoke speech recognition via AIDL
• Trailing Punctuation Unlimited Option: Added "Unlimited" option for trailing punctuation and emoji removal threshold

Improvements

• Soniox Model Upgrade: Upgraded Soniox non-streaming recognition model to v5
• Haptic Feedback Refactor: Refactored and centralized haptic feedback mechanism
• Recording Test Page Optimization: Optimized latency color display on the recording test page
• [Pro] AMOLED Independent Toggle: Refactored AMOLED mode as an independent toggle with automatic migration from old settings
• [Pro] Auto-Update Check Optimization: Optimized auto-update checks; when disabled, no longer depends on Play Services

Bug Fixes

• Keyboard Bottom Spacing: Optimized keyboard bottom spacing in certain scenarios
• ITN Optimization: Improved ITN performance in certain scenarios
• [Pro] Hotword Statistics Fix: Fixed hotword hit count statistics and color grading

📝 更新日志 / Changelog (v4.0.0)

中文

重大变更

• 设置页面 Compose 迁移: 将设置页面从传统 XML 布局全面迁移至 Jetpack Compose
• 键盘和悬浮球 UI 重构: 使用 Compose 重构键盘和悬浮球 UI，提升可维护性与渲染性能
• 自定义键盘布局：支持在画布中自定义键盘布局和内容
• [Pro]主题颜色系统重构: 重构主题颜色系统，添加 AMOLED 模式和多种 Monet 动态色支持

新增功能

• MiMo ASR 在线识别: 新增 MiMo 在线 ASR 供应商支持，多模态模型与 ASR 模型
• X-ASR 本地流式引擎: 引入 X-ASR 引擎替代 Paraformer，支持本地流式识别
• 音量键控制录音：支持借助无障碍权限实现音量键控制录音启停
• 录音测试与分析页面: 新增录音测试和分析页面
• 录音模式切换拓展按钮: 新增麦克风录音模式切换拓展按钮
• Miuix 组件库集成: 集成 Miuix 组件库以增强设置页 UI 体验
• 句末标点去除阈值: 添加去除句末标点与 emoji 的可配置字数阈值
• OpenAI completions 接口：OpenAI 语音识别渠道支持通过 completions 接口调用多模态模型进行转录
• [Pro]热词增强模式: 新增基于音素的热词增强模式，基于音素匹配动态替换识别结果
• [Pro]MiMo 上下文注入: 为 MiMo 多模态模型添加上下文注入支持
• [Pro]正则后处理顺序选项: 添加正则表达式在 AI 后处理前执行的选项，支持功能说明对话框

改进优化

• 本地 ITN 算法重构: 重构本地模型中文 ITN 模块，覆盖更多使用场景
• 升级 sherpa-onnx 至 1.13.2
• 为本地模型实现哈希完整性检查
• 改进大量组件 UI 细节和文案
• 优化无障碍权限检测和处理

问题修复

• 修复 mimo-v2.5 请求体构建和响应解析
• 优化无障碍事件处理与设置界面重组性能

English

Major Changes

• Settings Compose Migration: Migrated all settings pages from traditional XML layouts to Jetpack Compose
• Keyboard and Floating Ball UI Refactor: Refactored keyboard and floating ball UI using Compose for improved maintainability and rendering performance
• Custom Keyboard Layout: Support for customizing keyboard layout and content on canvas
• [Pro] Theme Color System Refactor: Refactored the theme color system with added AMOLED mode and multiple Monet dynamic color support

New Features

• MiMo ASR Online Recognition: Added MiMo online ASR vendor support with multimodal and ASR models
• X-ASR Local Streaming Engine: Introduced X-ASR engine to replace Paraformer, supporting local streaming recognition
• Volume Key Recording Control: Support for volume key recording start/stop via accessibility permission
• Recording Test and Analysis Page: Added recording test and analysis page
• Recording Mode Switch Extension Button: Added microphone recording mode switch extension button
• Miuix Library Integration: Integrated the Miuix component library to enhance settings page UI experience
• Trailing Punctuation Removal Threshold: Added configurable character count threshold for removing trailing punctuation and emoji
• OpenAI Completions API: OpenAI ASR channel supports calling multimodal models via the completions interface for transcription
• [Pro] Hotword Enhancement Mode: Added phoneme-based hotword enhancement mode, dynamically replacing recognition results based on phoneme matching
• [Pro] MiMo Context Injection: Added context injection support for MiMo multimodal models
• [Pro] Regex Post-Processing Order Option: Added option to run regex before AI post-processing, with feature description dialog

Improvements

• Local ITN Algorithm Refactoring: Refactored the local model Chinese ITN module, covering more use cases
• Upgraded sherpa-onnx to 1.13.2
• Implemented hash integrity check for local models
• Improved numerous component UI details and copy
• Optimized accessibility permission detection and handling

Bug Fixes

• Fixed mimo-v2.5 request body construction and response parsing
• Optimized accessibility event processing and settings page recomposition performance

📝 更新日志 / Changelog (v3.16.1)

中文

新增功能

• [Pro]输入框上下文: 在主键盘中使用时，LLM 后处理可获取输入法输入框现有文本作为上下文参考，提升改写的连贯性
• API 日志页本地 ASR 记录: API 调用日志页新增本地 ASR 调用记录支持

改进优化

• 空音频过滤默认关闭: 将 VAD 空音频自动取消与静音片段过滤功能默认关闭，用户可根据需求启用
• 空音频跳过提示优化: 优化空音频跳过时的错误提示文案，区分不同跳过原因
• 音频过滤器重写: 重写音频过滤器为基于分块的逐段分析与修剪，提升静音检测准确性
• 本地模型加载优化: 将启动录音时同步加载本地模型的功能覆盖到外部联动场景

问题修复

• 滑动锁定触摸事件处理: 修复麦克风按钮在滑动锁定模式下触摸事件的响应处理

English

New Features

• [Pro] Input Field Context: When using the main keyboard, LLM post-processing can now use the existing text in the input field as context, improving rewrite coherence
• API Log Page Local ASR Records: Added local ASR call logging support to the API Log page

Improvements

• Empty Audio Filtering Disabled by Default: Changed VAD empty audio auto-cancel and silent segment filtering to be disabled by default; users can enable it as needed
• Empty Audio Skip Message Optimization: Improved error messages for empty audio skips to differentiate between skip reasons
• Audio Filter Rewrite: Rewrote the audio filter with chunk-based analysis and trimming for improved silence detection accuracy
• Local Model Loading Optimization: Extended preload-on-recording logic to cover external linkage scenarios

Bug Fixes

• Swipe Lock Touch Event Handling: Fixed touch event handling for the microphone button when in swipe-locked mode

📝 更新日志 / Changelog (v3.16.0)

中文

新增功能

• OpenRouter ASR 支持: 新增 OpenRouter 在线 ASR 供应商支持，支持 API Key 与模型配置
• API 调用日志: 识别记录页新增 API 调用记录与查看页面，支持列表、筛选、搜索和详情查看
• VAD 空音频过滤: 新增非流式识别前过滤空音频请求与去除静音片段过滤的设置,默认开启
• 关于页识别统计: 关于页新增近 30 天在线 ASR 渠道失败率统计

改进优化

• 新增 armv7a 打包: 支持 armv7a ,并可按设备 ABI 选择安装包并增加备用下载地址，提升应用内更新下载兼容性
• [Pro]正则编辑输入框: 将正则规则编辑界面的输入框改为常规文本输入框

问题修复

• 识别取消处理: 在主键盘使用时,可以在后处理完成后一段时间撤回后处理的处理

English

New Features

• OpenRouter ASR Support: Added OpenRouter as an online ASR provider with API key and model configuration
• API Call Logs: Added API call recording and viewing to the recognition history page, with list, filter, search, and detail views
• VAD Empty Audio Filtering: Added settings to filter empty audio requests and remove silent segments before non-streaming recognition. Default enabled.
• About Page Recognition Stats: Added online ASR channel failure rate stats for the last 30 days on the About page

Improvements

• armv7a Packaging: Added armv7a support, selected APKs by device ABI, and added fallback download URLs to improve in-app update download compatibility
• [Pro] Regex Edit Input: Changed the input field in the regex rule editor to a regular text input field

Bug Fixes

• Recognition Cancellation Handling: When using the main keyboard, post-processing changes can be reverted for a short period after post-processing completes

📝 更新日志 / Changelog (v3.15.2)

中文

新增功能

• 非流式上传音频压缩: 新增在线非流式识别上传音频压缩选项，在服务商支持时减少传输体积和网络耗时，默认开启

改进优化

• Gemini 3 思考配置支持: Gemini ASR 增加 Gemini 3 模型的思考配置支持，并将相关设置文案调整为降低思考强度

English

New Features

• Non-Streaming Upload Audio Compression: Added an option to compress upload audio for online non-streaming recognition when supported by the selected provider, reducing transfer size and network time. Default enabled.

Improvements

• Gemini 3 Thinking Config Support: Added Gemini 3 model thinking configuration support for Gemini ASR and updated the related setting copy to reduce thinking intensity.

📝 更新日志 / Changelog (v3.15.1)

中文

问题修复

• 前台服务稳定性优化: 前台服务类型与异常处理补强，优化安卓15以上用户体验

English

问题修复

• Foreground Service Stability Optimization: Enhanced foreground service types and exception handling, improving user experience on Android 15 and above.

📝 更新日志 / Changelog (v3.15.0)

中文

新增功能

• Qwen3-ASR 本地识别支持: 新增 Qwen3-ASR 0.6B 本地 ASR 供应商
• Parakeet 本地识别支持: 新增 Parakeet 本地 ASR 引擎支持，V3版本支持多种欧洲语言，V2版本支持英语
• StepAudio ASR 支持: 新增 StepAudio ASR 在线供应商支持

改进优化

• 本地模型自动卸载时机: 调整预热启动成功后的自动卸载调度，改善本地模型资源管理
• [Pro]热词支持扩展: 为 Qwen3ASR、FunASR Nano、StepAudio 增加热词功能

问题修复

• 识别供应商标识修正: 优化 Qwen3Asr、FireRedAsr 和 Paraformer 的供应商标签
• 识别时长统计修复: 修复部分 ASR 供应商的识别时长上报问题

English

New Features

• Qwen3-ASR Local Recognition: Added Qwen3-ASR 0.6B as a new local ASR vendor
• Parakeet Local Recognition: Added support for the Parakeet local ASR engine, with V3 supporting multiple European languages and V2 supporting English
• StepAudio ASR Support: Added support for StepAudio as an online ASR vendor

Improvements

• Local Model Auto-Unload Timing: Adjusted auto-unload scheduling after warm-up starts successfully to improve local model resource management
• [Pro] Expanded Hotword Support: Added hotword support for Qwen3ASR, FunASR Nano, and StepAudio

Bug Fixes

• Vendor Tag Correction: Improved vendor tags for Qwen3Asr, FireRedAsr, and Paraformer
• Recognition Duration Reporting: Fixed duration reporting for some ASR vendors

📝 更新日志 / Changelog (v3.14.2)

中文

新增功能

• 阿里云百炼 Qwen3.5-Omni: 为阿里云百炼 ASR 新增 Qwen3.5-Omni Flash/Plus 模型选项

改进优化

• 录音启动网络预热: 在录音启动阶段增加网络预热,尝试降低首次使用的网络延迟
• 本地模型超时策略优化: 为不同本地识别引擎根据资源消耗设置不同的超时阈值
• 保活常驻通知内容优化: 常驻通知中支持显示实时刷新的基本信息

English

New Features

• Alibaba DashScope Qwen3.5-Omni: Added Qwen3.5-Omni Flash and Plus model options for Alibaba DashScope ASR

Improvements

• Recording Startup Network Warmup: Added network warmup during recording startup to help reduce first-use network latency
• Local Model Timeout Tuning: Set different timeout thresholds for different local ASR engines based on resource usage
• Keep-Alive Notification Content: Added support for showing live-updating basic information in the persistent notification

📝 更新日志 / Changelog (v3.14.1)

中文

新增功能

• OpenAI ASR 多配置支持: 为 OpenAI ASR 添加多配置切换支持
• FunASR Nano 语言选择: 为 FunASR Nano 添加识别语言选择功能
• FunASR MLT Nano 模型支持: 新增 FunASR MLT Nano 变体支持

改进优化

• FunASR Nano 中文处理优化: 启用原生 ITN，改善中文数字等文本格式化效果

English

New Features

• OpenAI ASR Multi-Profile Support: Added multi-profile switching support for OpenAI ASR
• FunASR Nano Language Selection: Added recognition language selection for FunASR Nano
• FunASR MLT Nano Variant Support: Added support for the FunASR MLT Nano variant

Improvements

• FunASR Nano Native ITN: Enabled native ITN in FunASR Nano to improve Chinese text normalization results

📝 更新日志 / Changelog (v3.14.0)

中文

新增功能

• FireRedASR V2 引擎支持: 替换 TeleSpeech 并新增 FireRedASR V2 引擎支持

改进优化

• 识别性能优化: 优化本地模型发现与 PCM 转换流程，减少冗余处理开销
• 界面渲染优化: 减少冗余 UI 更新，提升界面渲染效率
• 本地识别依赖升级: 升级 sherpa-onnx 至 1.12.28

问题修复

• 修复悬浮球下录音状态清理问题，提升稳定性

English

New Features

• FireRedASR V2 Engine Support: Replaced TeleSpeech and added support for the FireRedASR V2 engine

Improvements

• Recognition Performance Optimization: Optimized model discovery and PCM conversion flow to reduce redundant processing overhead
• UI Rendering Optimization: Reduced redundant UI updates to improve rendering efficiency
• Local ASR Dependency Upgrade: Upgraded sherpa-onnx to 1.12.28

Bug Fixes

• Fixed FireRedASR result sanitization and resource cleanup issues on error paths to improve recognition stability
• Fixed delayed UI updates caused by structural render refresh not being flushed in some scenarios