fix: 适配新版千问录音状态并升级至 1.0.3

AI-assisted-by: Codex
AI-model: gpt-5.6-sol
AI-reasoning-effort: medium
Co-authored-by: Codex <codex@openai.com>

AGENTS.md

@@ -2,7 +2,7 @@
 
 ## 项目范围
 
-TamaASR 是仅面向 Linux 的 Fcitx5 桌面语音输入工具。项目名为 `TamaASR`，分发包名为 `tama-asr`，当前版本为 `1.0.0`。
+TamaASR 是仅面向 Linux 的 Fcitx5 桌面语音输入工具。项目名为 `TamaASR`，分发包名为 `tama-asr`，当前版本为 `1.0.3`。
 
 不得恢复 Windows、PowerShell 或 Tauri 输出路径。发行版差异只能存在于依赖说明和打包配置，不能进入语音输入核心逻辑。
 

CHANGELOG.md

@@ -2,6 +2,19 @@
 
 本项目遵循 [Keep a Changelog](https://keepachangelog.com/zh-CN/1.1.0/) 的格式。
 
+## [1.0.3] - 2026-10-07
+
+### 修复
+
+- 适配新版千问网页的录音状态标记，修复松开快捷键后未点击停止按钮、悬浮球持续停留在黄色终止状态的问题。
+- Nix 包启动时显式使用 XWayland，与其他发行版的启动参数保持一致。
+
+## [1.0.2] - 2026-09-04
+
+### 修复
+
+- 修复快速松开快捷键时，网页尚未完成异步启动便再次点击麦克风，导致外部会话已结束但网页仍在录音的问题。
+
 ## [1.0.1] - 2026-09-01
 
 ### 变更

CLAUDE.md

@@ -2,7 +2,7 @@
 
 ## 项目范围
 
-TamaASR 是仅面向 Linux 的 Fcitx5 桌面语音输入工具。项目名为 `TamaASR`，分发包名为 `tama-asr`，当前版本为 `1.0.0`。
+TamaASR 是仅面向 Linux 的 Fcitx5 桌面语音输入工具。项目名为 `TamaASR`，分发包名为 `tama-asr`，当前版本为 `1.0.3`。
 
 不得恢复 Windows、PowerShell 或 Tauri 输出路径。发行版差异只能存在于依赖说明和打包配置，不能进入语音输入核心逻辑。
 

flake.lock

@@ -2,11 +2,11 @@
   "nodes": {
     "nixpkgs": {
       "locked": {
-        "lastModified": 1788107362,
-        "narHash": "sha256-pKxqcA71wmW6eMh3kFpW5NHIdbrgYOhtW2aOwqQ28GQ=",
+        "lastModified": 1790707128,
+        "narHash": "sha256-IB1Qi718DOScdq/skqHOKIApsAznqa8TkdaWX/MGh4w=",
         "owner": "Melorise",
         "repo": "nixpkgs",
-        "rev": "e1c33116d631fcc7987a1a6b298622523ce1d66a",
+        "rev": "521f999a323ecd2bee19a0e93d133c27712cbee1",
         "type": "github"
       },
       "original": {

linux/fcitx5/CMakeLists.txt

@@ -1,6 +1,6 @@
 cmake_minimum_required(VERSION 3.16)
 
-project(tama-asr-fcitx5 VERSION 1.0.1 LANGUAGES CXX)
+project(tama-asr-fcitx5 VERSION 1.0.3 LANGUAGES CXX)
 
 include(GNUInstallDirs)
 include(CTest)

linux/fcitx5/tama-asr.conf.in

@@ -6,7 +6,7 @@ Comment[zh_CN]=将 TamaASR 网页后端连接到当前输入上下文
 Type=SharedLibrary
 Library=@TAMA_ASR_LIBRARY_PATH@
 Category=Module
-Version=1.0.1
+Version=1.0.3
 OnDemand=False
 Configurable=False
 

package.json

@@ -1,7 +1,7 @@
 {
   "name": "tama-asr",
   "productName": "TamaASR",
-  "version": "1.0.1",
+  "version": "1.0.3",
   "private": true,
   "packageManager": "pnpm@11.21.0",
   "author": "TamaASR contributors",

packaging/README.md

@@ -1,13 +1,13 @@
 # Linux 分发打包
 
-TamaASR 1.0.1 的系统包由两部分组成：
+TamaASR 1.0.3 的系统包由两部分组成：
 
 - Electron 应用安装到 `/opt/tama-asr`，命令入口为 `/usr/bin/tama-asr`；
 - Fcitx5 常驻 addon 由目标发行版中的 CMake/Fcitx5 开发包编译，并安装到该环境报告的 addon 目录。
 
 安装包同时提供桌面入口、项目图标和 `/etc/xdg/autostart/tama-asr.desktop`，因此安装后的默认行为是随桌面会话启动。设置页关闭自启动时，应用需要在用户级 autostart 目录写入覆盖项，不能删除系统包所属文件。
 
-`tama-asr` 1.0.1 的 deb、RPM 和 Arch 元数据均声明替换旧 `meloasr` 包；通过对应发行版的包管理器安装新包时，旧应用、Fcitx5 addon 和桌面文件会一并卸载。
+`tama-asr` 1.0.3 的 deb、RPM 和 Arch 元数据均声明替换旧 `meloasr` 包；通过对应发行版的包管理器安装新包时，旧应用、Fcitx5 addon 和桌面文件会一并卸载。
 
 ## 统一构建契约
 
@@ -42,7 +42,7 @@ pnpm run build:linux:dir
 packaging/deb/build-deb.sh
 ```
 
-产物位于 `packaging/out/tama-asr_1.0.1_<arch>.deb`。依赖名称按 Debian/Ubuntu 系列填写，但尚未在各发行版完成安装验证。
+产物位于 `packaging/out/tama-asr_1.0.3_<arch>.deb`。依赖名称按 Debian/Ubuntu 系列填写，但尚未在各发行版完成安装验证。
 
 ## rpm
 

packaging/arch/PKGBUILD

@@ -1,5 +1,5 @@
 pkgname=tama-asr
-pkgver=1.0.1
+pkgver=1.0.3
 pkgrel=1
 pkgdesc='Web-backed voice input for Fcitx5'
 arch=('x86_64' 'aarch64')

packaging/arch/PKGBUILD-from-source

@@ -1,5 +1,5 @@
 pkgname=tama-asr
-pkgver=1.0.1
+pkgver=1.0.3
 pkgrel=1
 pkgdesc='Web-backed voice input for Fcitx5'
 arch=('x86_64')

packaging/arch/README.md

@@ -8,7 +8,7 @@
 
 ```bash
 curl --fail --location --remote-name \
-  https://github.com/Melorise/TamaASR/releases/download/v1.0.1/PKGBUILD
+  https://github.com/Melorise/TamaASR/releases/download/v1.0.3/PKGBUILD
 makepkg --syncdeps --cleanbuild
 ```
 
@@ -20,7 +20,7 @@ makepkg --syncdeps --cleanbuild
 
 ```bash
 curl --fail --location --remote-name \
-  https://github.com/Melorise/TamaASR/releases/download/v1.0.1/PKGBUILD-from-source
+  https://github.com/Melorise/TamaASR/releases/download/v1.0.3/PKGBUILD-from-source
 makepkg --syncdeps --cleanbuild -p PKGBUILD-from-source
 ```
 

packaging/assets/tama-asr.metainfo.xml

@@ -16,6 +16,8 @@
     <binary>tama-asr</binary>
   </provides>
   <releases>
+    <release version="1.0.3" date="2026-10-07"/>
+    <release version="1.0.2" date="2026-09-04"/>
     <release version="1.0.1" date="2026-09-01"/>
     <release version="1.0.0" date="2026-09-01"/>
     <release version="0.1.17" date="2026-09-01"/>

packaging/deb/build-deb.sh

@@ -2,7 +2,7 @@
 set -euo pipefail
 
 project_dir="$(cd -- "$(dirname -- "${BASH_SOURCE[0]}")/../.." && pwd)"
-version="${TAMA_ASR_VERSION:-1.0.1}"
+version="${TAMA_ASR_VERSION:-1.0.3}"
 arch="${TAMA_ASR_DEB_ARCH:-$(dpkg --print-architecture)}"
 app_dir="${1:-${TAMA_ASR_ELECTRON_APP_DIR:-${project_dir}/release/linux-unpacked}}"
 work_dir="${project_dir}/packaging/out/deb-root"

packaging/nix/package.nix

@@ -4,7 +4,7 @@
 , nodejs
 , pnpm_11
 , pnpmConfigHook
-, electron
+, electron_42
 , cmake
 , pkg-config
 , makeWrapper
@@ -84,7 +84,8 @@ stdenv.mkDerivation (finalAttrs: {
       "$out/share/pixmaps/tama-asr.png"
 
     cmake --install build/fcitx5
-    makeWrapper ${electron}/bin/electron "$out/bin/tama-asr" \
+    makeWrapper ${electron_42}/bin/electron "$out/bin/tama-asr" \
+      --add-flags "--ozone-platform=x11" \
       --add-flags "$appDir"
 
     runHook postInstall

packaging/rpm/build-rpm.sh

@@ -2,7 +2,7 @@
 set -euo pipefail
 
 project_dir="$(cd -- "$(dirname -- "${BASH_SOURCE[0]}")/../.." && pwd)"
-version="${TAMA_ASR_VERSION:-1.0.1}"
+version="${TAMA_ASR_VERSION:-1.0.3}"
 app_dir="${1:-${TAMA_ASR_ELECTRON_APP_DIR:-${project_dir}/release/linux-unpacked}}"
 stage_dir="${project_dir}/packaging/out/rpm-root"
 top_dir="${project_dir}/packaging/out/rpmbuild"

packaging/rpm/tama-asr-binary.spec

@@ -38,6 +38,13 @@ tar -C %{buildroot} -xzf %{SOURCE0}
 /etc/xdg/autostart/tama-asr.desktop
 
 %changelog
+* Wed Oct 07 2026 TamaASR contributors - 1.0.3-1
+- 适配新版千问录音状态，修复松开快捷键后无法终止的问题。
+- Nix 启动参数与其他发行版保持一致，显式使用 XWayland。
+
+* Fri Sep 04 2026 TamaASR contributors - 1.0.2-1
+- 修复快速停止可能使网页录音状态与外部会话不同步的问题。
+
 * Tue Sep 01 2026 TamaASR contributors - 1.0.1-1
 - Redesign the settings and overlay-position interfaces.
 

src/backends/qianwen.ts

@@ -21,6 +21,7 @@ export const qianwenAdapter: BackendWebAdapter = {
   isRecording(document) {
     const microphone = this.findMicrophone(document);
     return microphone?.getAttribute('aria-pressed') === 'true' ||
+      microphone?.classList.contains('bg-tag') === true ||
       /停止语音|stop\s*(voice|recording|input)/i.test(microphone ? controlDescription(microphone) : '');
   },
 

src/preload/backend.ts

@@ -1,7 +1,7 @@
 import { ipcRenderer } from 'electron';
 import type { BackendControlCommand, BackendPageStatus, BackendTranscriptPayload } from '../backends/contracts';
 import { getBackendWebAdapter } from '../backends/web-registry';
-import { shouldClickMicrophoneOnCancel } from './speech-session-state';
+import { shouldClickMicrophoneOnCancel, shouldClickMicrophoneOnStop } from './speech-session-state';
 import { shouldFinishAfterStop } from './stop-completion';
 
 const backendId = process.argv
@@ -180,9 +180,7 @@ function stopSpeech(): void {
   if (!active || stopping || !adapter) return;
   stopping = true;
   const stoppedAt = Date.now();
-  const wasRecordingAtStop = adapter.isRecording(document);
-  adapter.findMicrophone(document)?.click();
-  diagnose('stop-clicked', { microphone: controlSnapshot(adapter.findMicrophone(document)) });
+  let stopClicked = false;
   const finish = (): void => {
     emitCurrentEditorText();
     resetSession();
@@ -192,11 +190,22 @@ function stopSpeech(): void {
   const checkCompletion = (): void => {
     if (!active || !stopping || !adapter) return;
     const now = Date.now();
+    const isRecording = adapter.isRecording(document);
+    if (shouldClickMicrophoneOnStop(isRecording, stopClicked)) {
+      const microphone = adapter.findMicrophone(document);
+      if (microphone) {
+        microphone.click();
+        stopClicked = true;
+        diagnose('stop-clicked', { microphone: controlSnapshot(adapter.findMicrophone(document)) });
+      }
+      stopTimer = setTimeout(checkCompletion, 75);
+      return;
+    }
     if (shouldFinishAfterStop({
       elapsedMs: now - stoppedAt,
       quietMs: now - lastTranscriptAt,
-      wasRecordingAtStop,
-      isRecording: adapter.isRecording(document),
+      stopClicked,
+      isRecording,
       ...adapter.stopCompletion
     })) {
       diagnose('stop-complete', { elapsedMs: now - stoppedAt });
@@ -206,7 +215,7 @@ function stopSpeech(): void {
     stopTimer = setTimeout(checkCompletion, 75);
   };
   clearTimeout(stopTimer);
-  stopTimer = setTimeout(checkCompletion, 75);
+  checkCompletion();
 }
 
 function cancelSpeech(): void {

src/preload/speech-session-state.ts

@@ -1,3 +1,7 @@
 export function shouldClickMicrophoneOnCancel(active: boolean, stopping: boolean): boolean {
   return active && !stopping;
 }
+
+export function shouldClickMicrophoneOnStop(isRecording: boolean, stopClicked: boolean): boolean {
+  return isRecording && !stopClicked;
+}

src/preload/stop-completion.ts

@@ -1,7 +1,7 @@
 export interface StopCompletionInput {
   readonly elapsedMs: number;
   readonly quietMs: number;
-  readonly wasRecordingAtStop: boolean;
+  readonly stopClicked: boolean;
   readonly isRecording: boolean;
   readonly minimumWaitMs: number;
   readonly quietWindowMs: number;
@@ -9,7 +9,7 @@ export interface StopCompletionInput {
 }
 
 export function shouldFinishAfterStop(input: StopCompletionInput): boolean {
+  if (!input.stopClicked || input.isRecording) return false;
   if (input.elapsedMs >= input.timeoutMs) return true;
-  if (input.wasRecordingAtStop && input.isRecording) return false;
   return input.elapsedMs >= input.minimumWaitMs && input.quietMs >= input.quietWindowMs;
 }

test/qianwen.test.ts

@@ -0,0 +1,37 @@
+import assert from 'node:assert/strict';
+import test from 'node:test';
+import { qianwenAdapter } from '../src/backends/qianwen';
+import { shouldClickMicrophoneOnStop } from '../src/preload/speech-session-state';
+
+function microphone(className: string, attributes: Record<string, string | undefined> = {}): HTMLElement {
+  return {
+    classList: { contains: (token: string) => className.split(/\s+/).includes(token) },
+    getAttribute: (name: string) => name === 'class' ? className : attributes[name] ?? null,
+    querySelector: () => null,
+    textContent: ''
+  } as unknown as HTMLElement;
+}
+
+test('千问新版录音样式允许点击停止，恢复空闲样式后不再判为录音', () => {
+  let control = microphone('cursor-pointer hover:bg-tag');
+  const adapter = { ...qianwenAdapter, findMicrophone: () => control };
+  const document = {} as Document;
+  assert.equal(adapter.isRecording(document), false);
+
+  control = microphone('pointer-events-auto cursor-pointer bg-tag');
+  assert.equal(adapter.isRecording(document), true);
+  assert.equal(shouldClickMicrophoneOnStop(adapter.isRecording(document), false), true);
+  assert.equal(shouldClickMicrophoneOnStop(adapter.isRecording(document), true), false);
+
+  control = microphone('cursor-pointer hover:bg-tag');
+  assert.equal(adapter.isRecording(document), false);
+});
+
+test('千问旧版 aria 和停止文案保持兼容', () => {
+  for (const attributes of [{ 'aria-pressed': 'true' }, { 'aria-label': '停止语音' }]) {
+    const adapter = { ...qianwenAdapter, findMicrophone: () => microphone('', attributes) };
+    assert.equal(adapter.isRecording({} as Document), true);
+  }
+  const adapter = { ...qianwenAdapter, findMicrophone: () => null };
+  assert.equal(adapter.isRecording({} as Document), false);
+});

test/speech-session-state.test.ts

@@ -1,6 +1,6 @@
 import assert from 'node:assert/strict';
 import test from 'node:test';
-import { shouldClickMicrophoneOnCancel } from '../src/preload/speech-session-state';
+import { shouldClickMicrophoneOnCancel, shouldClickMicrophoneOnStop } from '../src/preload/speech-session-state';
 
 test('停止命令已经点击过麦克风后，取消不得再次点击并重新开启录音', () => {
   assert.equal(shouldClickMicrophoneOnCancel(true, true), false);
@@ -13,3 +13,12 @@ test('录音期间直接取消时仍应点击麦克风停止网页录音', () =>
 test('本地会话未激活时取消不得点击麦克风', () => {
   assert.equal(shouldClickMicrophoneOnCancel(false, false), false);
 });
+
+test('停止到达但网页尚未开始录音时不得再次点击开始按钮', () => {
+  assert.equal(shouldClickMicrophoneOnStop(false, false), false);
+});
+
+test('网页异步进入录音后只点击一次停止按钮', () => {
+  assert.equal(shouldClickMicrophoneOnStop(true, false), true);
+  assert.equal(shouldClickMicrophoneOnStop(true, true), false);
+});

test/stop-completion.test.ts

@@ -5,7 +5,7 @@ import { shouldFinishAfterStop } from '../src/preload/stop-completion';
 const base = {
   elapsedMs: 700,
   quietMs: 500,
-  wasRecordingAtStop: true,
+  stopClicked: true,
   isRecording: false,
   minimumWaitMs: 250,
   quietWindowMs: 450,
@@ -16,6 +16,10 @@ test('网页仍在录音时不得提前提交', () => {
   assert.equal(shouldFinishAfterStop({ ...base, isRecording: true }), false);
 });
 
+test('尚未点击停止时不得提交', () => {
+  assert.equal(shouldFinishAfterStop({ ...base, stopClicked: false }), false);
+});
+
 test('网页结束且最终文本静默窗口满足时立即提交', () => {
   assert.equal(shouldFinishAfterStop(base), true);
 });
@@ -24,6 +28,10 @@ test('网页结束后仍有文本修正时继续等待', () => {
   assert.equal(shouldFinishAfterStop({ ...base, quietMs: 120 }), false);
 });
 
-test('网页状态未回切时由超时兜底结束', () => {
-  assert.equal(shouldFinishAfterStop({ ...base, elapsedMs: 4_500, isRecording: true }), true);
+test('网页仍在录音时超时也不得提交', () => {
+  assert.equal(shouldFinishAfterStop({ ...base, elapsedMs: 4_500, isRecording: true }), false);
+});
+
+test('网页已停止时超时可以结束静默等待', () => {
+  assert.equal(shouldFinishAfterStop({ ...base, elapsedMs: 4_500, quietMs: 120 }), true);
 });