From 8652ee74c686f6d78ab61cf32d863f769b69ddfa Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?=E7=88=B1=E7=AC=91=E7=9A=84=E6=AF=9B=E6=AF=9B=E8=99=AB?=
 <315515767@qq.com>
Date: Fri, 7 Aug 2026 15:06:56 +0800
Subject: [PATCH] =?UTF-8?q?feat(web):=20=E5=AE=9E=E6=97=B6=E5=AF=B9?=
 =?UTF-8?q?=E8=AF=9D=20-=20=E7=AC=AC=E4=B8=80=E9=98=B6=E6=AE=B5=E8=BE=93?=
 =?UTF-8?q?=E5=85=A5=E8=A1=8C=E5=86=85=E8=81=94=E8=AF=AD=E9=9F=B3=E5=BD=95?=
 =?UTF-8?q?=E5=88=B6=E4=B8=8E=E6=B5=8F=E8=A7=88=E5=99=A8=20STT=20=E8=BD=AC?=
 =?UTF-8?q?=E5=86=99=20(#165)?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Co-authored-by: zhaoyue <zhaoyue@rongshantech.com>
---
 .../canvas/canvas-assistant-panel.tsx         |   5 +
 .../canvas/canvas-local-agent-panel.tsx       |  11 +-
 .../components/conversation/IMPLEMENTATION.md | 109 ++++++++++
 web/src/components/conversation/README.md     | 100 +++++++++
 .../conversation/audio-waveform.tsx           |  69 +++++++
 web/src/components/conversation/index.ts      |   3 +
 .../conversation/voice-recording-button.tsx   |  49 +++++
 .../conversation/voice-recording-inline.tsx   | 191 +++++++++++++++++
 web/src/hooks/use-speech-recognition.ts       | 163 +++++++++++++++
 web/src/hooks/use-voice-recording.ts          | 193 ++++++++++++++++++
 web/src/pages/create/index.tsx                |   5 +
 web/src/pages/test-voice-recording.tsx        | 107 ++++++++++
 web/src/router.tsx                            |   2 +
 13 files changed, 1006 insertions(+), 1 deletion(-)
 create mode 100644 web/src/components/conversation/IMPLEMENTATION.md
 create mode 100644 web/src/components/conversation/README.md
 create mode 100644 web/src/components/conversation/audio-waveform.tsx
 create mode 100644 web/src/components/conversation/index.ts
 create mode 100644 web/src/components/conversation/voice-recording-button.tsx
 create mode 100644 web/src/components/conversation/voice-recording-inline.tsx
 create mode 100644 web/src/hooks/use-speech-recognition.ts
 create mode 100644 web/src/hooks/use-voice-recording.ts
 create mode 100644 web/src/pages/test-voice-recording.tsx

diff --git a/web/src/components/canvas/canvas-assistant-panel.tsx b/web/src/components/canvas/canvas-assistant-panel.tsx
index fc09ee1da..7a16c8ba8 100644
--- a/web/src/components/canvas/canvas-assistant-panel.tsx
+++ b/web/src/components/canvas/canvas-assistant-panel.tsx
@@ -17,6 +17,7 @@ import { navigateToSettings } from "@/lib/settings-navigation";
 import { cinematicAgentSessionOpsJson, createCinematicAgentSession, isAgentSessionPollingAbort, resumeCinematicAgentSession } from "@/lib/canvas/canvas-agent-session";
 import { summarizeCanvasContext } from "@/lib/canvas/canvas-context-summary";
 import { AgentChatComposer, AgentChatMessage, AgentPanelTabs, AgentWorkingMessage, type CanvasAgentChatMessage, type CanvasAgentMode } from "./canvas-agent-chat-ui";
+import { VoiceRecordingButton } from "@/components/conversation/voice-recording-button";
 import { AgentChatEmptyState, AgentPanelChrome } from "./canvas-agent-panel-chrome";
 import { CanvasLocalAgentPanel } from "./canvas-local-agent-panel";
 import { NODE_DEFAULT_SIZE } from "@/constant/canvas";
@@ -781,6 +782,10 @@ export function CanvasAssistantPanel({ nodes, selectedNodeIds, snapshot, project
                         onAddFiles={addImagesToCanvas}
                         left={
                             <>
+                                <VoiceRecordingButton
+                                    disabled={agentBusy}
+                                    onTranscribed={(text) => setPrompt((prev) => (prev.trim() ? `${prev} ${text}` : text))}
+                                />
                                 <AgentTextModelPicker config={effectiveConfig} value={effectiveConfig.textModel} onChange={(model) => updateConfig("textModel", model)} />
                                 {cinematicEntryActive ? <span className="ml-2 inline-flex h-6 items-center rounded-md px-2 text-[var(--fs-tiny)] font-medium" style={{ background: theme.spatial.surface, color: theme.node.muted }}>影视项目</span> : null}
                             </>
diff --git a/web/src/components/canvas/canvas-local-agent-panel.tsx b/web/src/components/canvas/canvas-local-agent-panel.tsx
index 0dd2bb2fc..18fc8521d 100644
--- a/web/src/components/canvas/canvas-local-agent-panel.tsx
+++ b/web/src/components/canvas/canvas-local-agent-panel.tsx
@@ -13,6 +13,7 @@ import { useCanvasAgentStore, type AgentAttachment, type AgentChatItem, type Age
 import { previewCanvasAgentOps, summarizeCanvasAgentOps, type CanvasAgentOp, type CanvasAgentSnapshot } from "@/lib/canvas/canvas-agent-ops";
 import { isProjectAgentReadTool, isProjectAgentToolName, runProjectAgentTool } from "@/services/api/project-agent-tools";
 import { AgentChatComposer, AgentChatMessage, AgentPanelTabs, AgentPendingToolCard, AgentWorkingMessage, type CanvasAgentChatAttachment } from "./canvas-agent-chat-ui";
+import { VoiceRecordingButton } from "@/components/conversation/voice-recording-button";
 import { AgentChatEmptyState } from "./canvas-agent-panel-chrome";
 
 const PANEL_MOTION_SECONDS = 0.5;
@@ -562,7 +563,15 @@ export const CanvasLocalAgentPanel = memo(function CanvasLocalAgentPanel({ snaps
                         onSubmit={sendPrompt}
                         onAddFiles={addAttachments}
                         onRemoveAttachment={removeAttachment}
-                        left={attachments.length ? <span className="text-[var(--fs-label)]" style={{ color: theme.node.muted }}>{formatBytes(attachmentPayloadBytes(attachments))} / 30MB</span> : null}
+                        left={
+                            <>
+                                <VoiceRecordingButton
+                                    disabled={!connected || sending || waiting}
+                                    onTranscribed={(text) => setAgentState({ prompt: prompt.trim() ? `${prompt} ${text}` : text })}
+                                />
+                                {attachments.length ? <span className="text-[var(--fs-label)]" style={{ color: theme.node.muted }}>{formatBytes(attachmentPayloadBytes(attachments))} / 30MB</span> : null}
+                            </>
+                        }
                     />
                 </>
             )}
diff --git a/web/src/components/conversation/IMPLEMENTATION.md b/web/src/components/conversation/IMPLEMENTATION.md
new file mode 100644
index 000000000..75e93ebbc
--- /dev/null
+++ b/web/src/components/conversation/IMPLEMENTATION.md
@@ -0,0 +1,109 @@
+# 实时对话功能实现（第一阶段 MVP）
+
+## 项目概述
+
+实现第一阶段（MVP）：**本地语音录制 + 浏览器语音识别转写**。点击"实时对话"按钮 → 输入行内展开波形录制条 → 停止后自动转写为文字 → 文字填入输入框或直接发送。
+
+第一阶段**不接入语音大模型、不依赖后端与 API Key**，转写由浏览器内置 Web Speech API 完成，零配置可用。
+
+---
+
+## 交互流程
+
+1. 点击"实时对话"按钮（`VoiceRecordingButton`）
+2. 输入行内展开 `VoiceRecordingInline` 录制条（不弹窗），自动开始录音
+3. 显示波形动画（AnalyserNode 实时音量）+ 时长
+4. 点击停止 → 停止录音并触发浏览器语音识别
+5. 识别成功 → 短暂显示"转写完成" → 回调文本，由调用方填入输入框或直接发送
+6. 识别失败（无语音 / 权限拒绝 / 网络异常）→ 录制条内提示并支持重试
+
+---
+
+## 技术架构
+
+```
+用户点击"实时对话"按钮
+        │
+        ▼
+VoiceRecordingButton（输入行内展开录制条，不弹窗）
+        │
+        ▼
+VoiceRecordingInline 内联波形录制条
+        ├── useVoiceRecording：MediaRecorder 录制 + AnalyserNode 实时波形
+        ├── AudioWaveform：SVG 波形动画 + 时长显示
+        └── useSpeechRecognition：Web Speech API 实时转写
+                │
+                ▼
+点击停止 → speech.stop() 返回文本 → onTranscribed(文本) → 填入输入框
+```
+
+## 前端实现
+
+### 组件
+
+| 文件 | 说明 |
+|------|------|
+| `components/conversation/voice-recording-button.tsx` | 语音输入按钮：点击后输入行内展开录制条（局部状态，不弹窗） |
+| `components/conversation/voice-recording-inline.tsx` | 内联波形录制条：波形 + 时长 + 停止/取消，停止后自动转写 |
+| `components/conversation/audio-waveform.tsx` | SVG 波形可视化组件 |
+| `components/conversation/index.ts` | 组件统一导出 |
+
+### Hook
+
+| 文件 | 说明 |
+|------|------|
+| `hooks/use-voice-recording.ts` | 语音录制逻辑（MediaRecorder + AnalyserNode，波形/时长/停止/取消） |
+| `hooks/use-speech-recognition.ts` | 浏览器语音识别（Web Speech API，start/stop/cancel，返回文本与错误） |
+
+## 接入位置
+
+- 创建页主输入行：`web/src/pages/create/index.tsx`（`CreationComposer` 的 `creation-chat-controls`）
+- 画布助手面板：`canvas-assistant-panel.tsx`、`canvas-local-agent-panel.tsx`（输入行 left 插槽）
+- 测试页：`web/src/pages/test-voice-recording.tsx`（路由 `/test-voice-recording`）
+
+## 使用示例
+
+```tsx
+import { VoiceRecordingButton } from "@/components/conversation/voice-recording-button";
+
+<VoiceRecordingButton
+  disabled={isBusy}
+  onTranscribed={(text) => setPrompt((prev) => (prev.trim() ? `${prev} ${text}` : text))}
+/>
+```
+
+## 浏览器支持
+
+- 语音识别使用 Web Speech API（`SpeechRecognition` / `webkitSpeechRecognition`），Chrome / Edge 支持
+- 不支持时录制条内提示"当前浏览器不支持语音识别，请使用 Chrome 或 Edge 浏览器"，并可取消
+- 识别语言默认 `zh-CN`，可通过 `useSpeechRecognition({ lang })` 调整
+
+## 已删除内容
+
+以下第二阶段实验代码已全部删除：
+
+### 前端
+
+- `components/conversation/realtime-conversation.tsx`、`realtime-conversation-button.tsx`、`voice-recording-overlay.tsx`
+- `hooks/use-websocket.ts`、`stores/conversation-mode-store.ts`
+- `services/api/stt.ts`（后端 Whisper 转写客户端）
+
+### 后端
+
+- `service/stt.go`（OpenAI Whisper 转写）、`model/stt.go`（`stt_results` 表）
+- `handler/routes.go` 中 `RegisterSTTRoutes`、`main.go` 中路由注册
+- `database/schema.go` 中 `STTResult` 模型注册、`repository/repository.go` 中 `CreateSTTResult`
+
+## 验证情况
+
+- 后端 `go build ./...`：通过
+- 前端 `npx tsc --noEmit`：无本次新增类型错误（上游遗留 style profile / image.ts 错误除外）
+- 浏览器手动测试（真实麦克风 + 语音识别）未运行，需在 Chrome/Edge 下验证
+
+## 下一步
+
+- [x] 内联波形录制 + 浏览器语音识别自动转写（第一阶段 MVP）
+- [x] 移除第二阶段实验代码（WebSocket / WebRTC / Realtime API）
+- [ ] 语音播放功能（TTS）
+- [ ] 语音消息发送和显示
+- [ ] 真正实时对话（WebSocket + WebRTC + Realtime API，后续阶段）
diff --git a/web/src/components/conversation/README.md b/web/src/components/conversation/README.md
new file mode 100644
index 000000000..f2431b2c5
--- /dev/null
+++ b/web/src/components/conversation/README.md
@@ -0,0 +1,100 @@
+# 实时对话功能组件（第一阶段 MVP）
+本目录实现第一阶段（MVP）：**本地语音录制 + 浏览器语音识别转写**。点击"实时对话"按钮 → 输入行内展开波形录制条 → 停止后自动转写为文字 → 文字填入输入框或直接发送。
+
+## 零配置说明
+
+第一阶段**不接入语音大模型、不依赖后端与 API Key**：录音完成后直接使用浏览器内置 Web Speech API（`SpeechRecognition`）识别，转写全程在前端完成。
+
+- 支持的浏览器：Chrome / Edge（Chromium 内核）
+- 不支持时录制条内会明确提示，并可取消
+- 识别语言默认 `zh-CN`，见 `useSpeechRecognition` 的 `lang` 参数
+
+---
+
+## 组件列表
+
+### 1. `VoiceRecordingButton` - 语音输入按钮
+
+点击后输入行内展开波形录制条（不弹窗），自动开始录音；停止后自动转写并回调文本。使用局部状态，多个输入行可独立使用。
+
+**使用示例：**
+```tsx
+import { VoiceRecordingButton } from "@/components/conversation/voice-recording-button";
+
+<VoiceRecordingButton
+  disabled={isBusy}
+  onTranscribed={(text) => setPrompt((prev) => (prev.trim() ? `${prev} ${text}` : text))}
+/>
+```
+
+### 2. `VoiceRecordingInline` - 内联波形录制条
+
+输入行内展示的录制控件：波形动画 + 时长 + 停止/取消按钮；停止后自动转写，失败可重试。`VoiceRecordingButton` 默认使用本组件。
+
+**使用示例：**
+```tsx
+import { VoiceRecordingInline } from "@/components/conversation/voice-recording-inline";
+
+<VoiceRecordingInline
+  onTranscribed={(text) => setPrompt((prev) => (prev.trim() ? `${prev} ${text}` : text))}
+  onCancel={() => setOpen(false)}
+/>
+```
+
+### 3. `AudioWaveform` - 音频波形可视化
+
+SVG 波形可视化组件，支持实时数据更新。
+
+---
+
+## Hook
+
+### `useVoiceRecording` - 语音录制 Hook
+
+封装 MediaRecorder API 和 AnalyserNode，提供录制状态、波形数据、控制方法（供波形动画与时长展示）。
+
+### `useSpeechRecognition` - 浏览器语音识别 Hook
+
+封装 Web Speech API（`SpeechRecognition`/`webkitSpeechRecognition`），提供 `start()` / `stop(): Promise<string>` / `cancel()`，返回是否支持、识别错误。转写不依赖后端。
+
+---
+
+## 完整集成示例
+
+```tsx
+import { useState } from "react";
+import { VoiceRecordingButton } from "@/components/conversation/voice-recording-button";
+
+function ChatInput() {
+  const [prompt, setPrompt] = useState("");
+
+  const handleTranscribed = (text: string) => {
+    // 转写结果填入输入框，用户确认后发送
+    setPrompt((current) => (current.trim() ? `${current} ${text}` : text));
+  };
+
+  return (
+    <div className="flex items-center gap-2">
+      <textarea value={prompt} onChange={(e) => setPrompt(e.target.value)} placeholder="输入消息..." />
+      {/* 点击后输入行内展开波形录制条，停止后自动转写 */}
+      <VoiceRecordingButton onTranscribed={handleTranscribed} />
+      <button>发送</button>
+    </div>
+  );
+}
+```
+
+## 接入位置
+
+- 创建页主输入行：`web/src/pages/create/index.tsx`（`CreationComposer` 的 `creation-chat-controls`）
+- 画布助手面板：`canvas-assistant-panel.tsx`、`canvas-local-agent-panel.tsx`（输入行 left 插槽）
+- 测试页：`web/src/pages/test-voice-recording.tsx`
+
+## 下一步
+- [x] 内联波形录制（不弹窗），停止后自动转写
+- [x] 移除第二阶段实验代码（WebSocket / Realtime API / 旧浮层组件）
+- [x] 移除后端 STT 接口与数据库表（第一阶段不接入语音大模型）
+- [ ] 语音播放功能（TTS）
+- [ ] 语音消息发送和显示
+- [ ] 多模态对话（图片、视频）
+- [ ] 真正实时对话（WebSocket + WebRTC + Realtime API，后续阶段）
diff --git a/web/src/components/conversation/audio-waveform.tsx b/web/src/components/conversation/audio-waveform.tsx
new file mode 100644
index 000000000..31517e102
--- /dev/null
+++ b/web/src/components/conversation/audio-waveform.tsx
@@ -0,0 +1,69 @@
+import { useMemo } from "react";
+
+type AudioWaveformProps = {
+    waveform: number[];
+    /** 波形颜色，默认使用 CSS 变量 */
+    color?: string;
+    /** 波形高度（像素），默认 48 */
+    height?: number;
+    /** 波形宽度（像素），默认 200 */
+    width?: number;
+    /** 是否动画，默认 true */
+    animated?: boolean;
+};
+
+/**
+ * 音频波形可视化组件
+ * 使用 SVG 绘制实时波形
+ */
+export function AudioWaveform({
+    waveform,
+    color = "var(--cn-accent, #3b82f6)",
+    height = 48,
+    width = 200,
+    animated = true,
+}: AudioWaveformProps) {
+    // 如果没有波形数据，显示占位波形
+    const displayWaveform = useMemo(() => {
+        if (waveform.length === 0) {
+            // 生成静态占位波形
+            return Array.from({ length: 32 }, (_, i) => {
+                const x = i / 31;
+                return 0.3 + 0.2 * Math.sin(x * Math.PI * 2);
+            });
+        }
+        return waveform;
+    }, [waveform]);
+
+    const barWidth = width / displayWaveform.length;
+    const gap = Math.max(1, barWidth * 0.2);
+    const actualBarWidth = barWidth - gap;
+
+    return (
+        <svg
+            width={width}
+            height={height}
+            viewBox={`0 0 ${width} ${height}`}
+            className={animated ? "animate-pulse" : ""}
+            style={{ overflow: "visible" }}
+        >
+            {displayWaveform.map((value, index) => {
+                const barHeight = Math.max(2, value * height * 0.8);
+                const x = index * barWidth + gap / 2;
+                const y = (height - barHeight) / 2;
+                return (
+                    <rect
+                        key={index}
+                        x={x}
+                        y={y}
+                        width={actualBarWidth}
+                        height={barHeight}
+                        fill={color}
+                        rx={actualBarWidth / 2}
+                        opacity={0.6 + value * 0.4}
+                    />
+                );
+            })}
+        </svg>
+    );
+}
diff --git a/web/src/components/conversation/index.ts b/web/src/components/conversation/index.ts
new file mode 100644
index 000000000..5737c0149
--- /dev/null
+++ b/web/src/components/conversation/index.ts
@@ -0,0 +1,3 @@
+export { AudioWaveform } from "./audio-waveform";
+export { VoiceRecordingButton } from "./voice-recording-button";
+export { VoiceRecordingInline } from "./voice-recording-inline";
\ No newline at end of file
diff --git a/web/src/components/conversation/voice-recording-button.tsx b/web/src/components/conversation/voice-recording-button.tsx
new file mode 100644
index 000000000..f234d0477
--- /dev/null
+++ b/web/src/components/conversation/voice-recording-button.tsx
@@ -0,0 +1,49 @@
+import { useState } from "react";
+import { Button, Tooltip } from "antd";
+import { Mic } from "lucide-react";
+
+import { VoiceRecordingInline } from "./voice-recording-inline";
+import { canvasThemes } from "@/lib/canvas-theme";
+import { useThemeStore } from "@/stores/use-theme-store";
+
+type VoiceRecordingButtonProps = {
+    /** 转写完成回调，返回转写文本 */
+    onTranscribed: (text: string) => void;
+    /** 是否禁用（如发送中或未连接） */
+    disabled?: boolean;
+};
+
+/**
+ * 语音输入按钮：点击后在输入行内展开波形录制条，录制完成自动 STT 转写
+ * 使用局部状态，多个输入行可独立使用
+ */
+export function VoiceRecordingButton({ onTranscribed, disabled }: VoiceRecordingButtonProps) {
+    const theme = canvasThemes[useThemeStore((state) => state.theme)];
+    const [open, setOpen] = useState(false);
+
+    return (
+        <>
+            <Tooltip title="实时对话">
+                <Button
+                    type="text"
+                    shape="circle"
+                    className="!h-8 !w-8 !min-w-8"
+                    disabled={disabled}
+                    style={{ color: theme.node.muted }}
+                    icon={<Mic className="size-4" />}
+                    onClick={() => setOpen(true)}
+                    aria-label="实时对话"
+                />
+            </Tooltip>
+            {open ? (
+                <VoiceRecordingInline
+                    onTranscribed={(text) => {
+                        setOpen(false);
+                        onTranscribed(text);
+                    }}
+                    onCancel={() => setOpen(false)}
+                />
+            ) : null}
+        </>
+    );
+}
\ No newline at end of file
diff --git a/web/src/components/conversation/voice-recording-inline.tsx b/web/src/components/conversation/voice-recording-inline.tsx
new file mode 100644
index 000000000..9606d997a
--- /dev/null
+++ b/web/src/components/conversation/voice-recording-inline.tsx
@@ -0,0 +1,191 @@
+import { useEffect, useRef, useState } from "react";
+import { Button, Spin, Tooltip } from "antd";
+import { Check, Mic, Square, X } from "lucide-react";
+
+import { AudioWaveform } from "./audio-waveform";
+import { useSpeechRecognition } from "@/hooks/use-speech-recognition";
+import { useVoiceRecording } from "@/hooks/use-voice-recording";
+import { canvasThemes } from "@/lib/canvas-theme";
+import { useThemeStore } from "@/stores/use-theme-store";
+
+type VoiceRecordingInlineProps = {
+    /** 转写完成回调，返回转写文本 */
+    onTranscribed: (text: string) => void;
+    /** 取消录制回调 */
+    onCancel: () => void;
+};
+
+type TranscribeState = "idle" | "transcribing" | "done" | "error";
+
+/**
+ * 内联语音录制条（输入行内展示）
+ * 挂载后自动开始录音，显示波形动画；点击停止后自动转写为文字（浏览器 Web Speech API，无需后端与 API Key）
+ */
+export function VoiceRecordingInline({ onTranscribed, onCancel }: VoiceRecordingInlineProps) {
+    const theme = canvasThemes[useThemeStore((state) => state.theme)];
+    const {
+        state,
+        waveform,
+        duration,
+        error: recorderError,
+        start: startRecording,
+        stop: stopRecording,
+        cancel: cancelRecording,
+    } = useVoiceRecording({
+        maxDuration: 60,
+    });
+    const {
+        supported: speechSupported,
+        error: speechError,
+        start: startSpeech,
+        stop: stopSpeech,
+        cancel: cancelSpeech,
+    } = useSpeechRecognition();
+    const [transcribeState, setTranscribeState] = useState<TranscribeState>("idle");
+    const [transcribeError, setTranscribeError] = useState("");
+    const stopRequestedRef = useRef(false);
+    const wasRecordingRef = useRef(false);
+
+    // 挂载后自动开始录音与语音识别，卸载时清理
+    useEffect(() => {
+        if (!speechSupported) {
+            setTranscribeError("当前浏览器不支持语音识别，请使用 Chrome 或 Edge 浏览器");
+            setTranscribeState("error");
+            return;
+        }
+        void startRecording();
+        startSpeech();
+        return () => {
+            cancelRecording();
+            cancelSpeech();
+        };
+    }, [startRecording, cancelRecording, startSpeech, cancelSpeech, speechSupported]);
+
+    const handleStop = async () => {
+        if (stopRequestedRef.current || transcribeState !== "idle") return;
+        stopRequestedRef.current = true;
+        setTranscribeState("transcribing");
+        setTranscribeError("");
+        const text = await stopSpeech();
+        void stopRecording();
+        const trimmed = text.trim();
+        if (!trimmed) {
+            setTranscribeError("未识别到语音内容，请重试");
+            setTranscribeState("error");
+            stopRequestedRef.current = false;
+            return;
+        }
+        setTranscribeState("done");
+        // 短暂展示成功状态后回调
+        window.setTimeout(() => onTranscribed(trimmed), 600);
+    };
+
+    // 录音达到最大时长自动停止时，自动触发转写
+    useEffect(() => {
+        if (state === "recording") wasRecordingRef.current = true;
+        if (state === "idle" && wasRecordingRef.current && transcribeState === "idle" && !stopRequestedRef.current) {
+            wasRecordingRef.current = false;
+            void handleStop();
+        }
+    }, [state, transcribeState]);
+
+    const handleRetry = () => {
+        stopRequestedRef.current = false;
+        wasRecordingRef.current = false;
+        setTranscribeError("");
+        setTranscribeState("idle");
+        void startRecording();
+        startSpeech();
+    };
+
+    const formatDuration = (seconds: number) => {
+        const mins = Math.floor(seconds / 60);
+        const secs = Math.floor(seconds % 60);
+        return `${mins}:${secs.toString().padStart(2, "0")}`;
+    };
+
+    const displayError = recorderError || (speechError ? speechError.message : "") || (transcribeState === "error" ? transcribeError : "");
+
+    return (
+        <div
+            className="flex min-w-0 flex-1 items-center gap-2 rounded-lg border px-2 py-1.5"
+            style={{ background: theme.toolbar.panel, borderColor: theme.toolbar.border }}
+        >
+            {displayError ? (
+                <div className="flex min-w-0 flex-1 items-center gap-2">
+                    <span className="truncate text-xs" style={{ color: "#dc2626" }}>
+                        {displayError}
+                    </span>
+                    {speechSupported ? (
+                        <Tooltip title="重试">
+                            <Button
+                                type="text"
+                                size="small"
+                                icon={<Mic className="size-3.5" />}
+                                onClick={handleRetry}
+                                style={{ color: theme.node.muted }}
+                            />
+                        </Tooltip>
+                    ) : null}
+                    <Tooltip title="取消">
+                        <Button
+                            type="text"
+                            size="small"
+                            icon={<X className="size-3.5" />}
+                            onClick={onCancel}
+                            style={{ color: theme.node.muted }}
+                        />
+                    </Tooltip>
+                </div>
+            ) : transcribeState === "transcribing" ? (
+                <div className="flex items-center gap-2 px-2" style={{ color: theme.node.muted }}>
+                    <Spin size="small" />
+                    <span className="text-xs">正在转写...</span>
+                </div>
+            ) : transcribeState === "done" ? (
+                <div className="flex items-center gap-2 px-2" style={{ color: "#16a34a" }}>
+                    <Check className="size-4" />
+                    <span className="text-xs">转写完成</span>
+                </div>
+            ) : (
+                <>
+                    <AudioWaveform
+                        waveform={waveform}
+                        color={theme.accent.primary}
+                        height={32}
+                        width={160}
+                        animated={state === "recording"}
+                    />
+                    {state === "recording" ? (
+                        <span className="font-mono text-xs tabular-nums" style={{ color: theme.node.text }}>
+                            {formatDuration(duration)}
+                        </span>
+                    ) : null}
+                    <Tooltip title="取消">
+                        <Button
+                            type="text"
+                            size="small"
+                            icon={<X className="size-3.5" />}
+                            onClick={() => {
+                                cancelRecording();
+                                cancelSpeech();
+                                onCancel();
+                            }}
+                            style={{ color: theme.node.muted }}
+                        />
+                    </Tooltip>
+                    <Tooltip title="停止并转写">
+                        <Button
+                            type="text"
+                            size="small"
+                            icon={<Square className="size-3.5" />}
+                            onClick={handleStop}
+                            disabled={state !== "recording" || transcribeState !== "idle"}
+                            style={{ color: theme.accent.primary }}
+                        />
+                    </Tooltip>
+                </>
+            )}
+        </div>
+    );
+}
diff --git a/web/src/hooks/use-speech-recognition.ts b/web/src/hooks/use-speech-recognition.ts
new file mode 100644
index 000000000..37d2718ab
--- /dev/null
+++ b/web/src/hooks/use-speech-recognition.ts
@@ -0,0 +1,163 @@
+import { useCallback, useEffect, useRef, useState } from "react";
+
+type SpeechRecognitionAlternative = { transcript: string; confidence: number };
+type SpeechRecognitionResult = { isFinal: boolean; length: number; [index: number]: SpeechRecognitionAlternative };
+type SpeechRecognitionResultList = { length: number; item(index: number): SpeechRecognitionResult; [index: number]: SpeechRecognitionResult };
+
+type SpeechRecognitionLike = {
+    lang: string;
+    continuous: boolean;
+    interimResults: boolean;
+    maxAlternatives: number;
+    onstart: (() => void) | null;
+    onresult: ((event: { results: SpeechRecognitionResultList }) => void) | null;
+    onerror: ((event: { error: string }) => void) | null;
+    onend: (() => void) | null;
+    start: () => void;
+    stop: () => void;
+    abort: () => void;
+};
+
+type SpeechRecognitionConstructor = new () => SpeechRecognitionLike;
+
+function getSpeechRecognition(): SpeechRecognitionConstructor | null {
+    if (typeof window === "undefined") return null;
+    const globalWithSpeech = window as unknown as {
+        SpeechRecognition?: SpeechRecognitionConstructor;
+        webkitSpeechRecognition?: SpeechRecognitionConstructor;
+    };
+    return globalWithSpeech.SpeechRecognition || globalWithSpeech.webkitSpeechRecognition || null;
+}
+
+export type SpeechRecognitionError = {
+    code: string;
+    message: string;
+};
+
+type UseSpeechRecognitionOptions = {
+    /** 语言代码，默认 zh-CN */
+    lang?: string;
+};
+
+export type UseSpeechRecognitionReturn = {
+    /** 浏览器是否支持 Web Speech API */
+    supported: boolean;
+    /** 识别错误（权限、网络、不支持等） */
+    error: SpeechRecognitionError | null;
+    /** 开始识别，挂载后自动调用 */
+    start: () => void;
+    /** 停止识别并返回已累积的最终文本 */
+    stop: () => Promise<string>;
+    /** 取消识别并清空累积文本 */
+    cancel: () => void;
+};
+
+/**
+ * 浏览器内置语音识别 Hook（Web Speech API）
+ * 第一阶段 MVP 使用：不依赖后端、模型渠道或 API Key，零配置可用
+ */
+export function useSpeechRecognition(options: UseSpeechRecognitionOptions = {}): UseSpeechRecognitionReturn {
+    const { lang = "zh-CN" } = options;
+    const recognitionRef = useRef<SpeechRecognitionLike | null>(null);
+    const finalTextRef = useRef("");
+    const stopResolveRef = useRef<((text: string) => void) | null>(null);
+    const stopTimerRef = useRef<number | null>(null);
+    const [supported] = useState(() => Boolean(getSpeechRecognition()));
+    const [error, setError] = useState<SpeechRecognitionError | null>(null);
+
+    const start = useCallback(() => {
+        const Constructor = getSpeechRecognition();
+        if (!Constructor) {
+            setError({ code: "unsupported", message: "当前浏览器不支持语音识别，请使用 Chrome 或 Edge" });
+            return;
+        }
+        finalTextRef.current = "";
+        const recognition = new Constructor();
+        recognition.lang = lang;
+        recognition.continuous = true;
+        recognition.interimResults = false;
+        recognition.maxAlternatives = 1;
+        recognition.onstart = () => setError(null);
+        recognition.onresult = (event) => {
+            let text = finalTextRef.current;
+            for (let i = 0; i < event.results.length; i += 1) {
+                const result = event.results[i];
+                if (result.isFinal && result[0]?.transcript) {
+                    text = [text, result[0].transcript.trim()].filter(Boolean).join(" ");
+                }
+            }
+            finalTextRef.current = text;
+        };
+        recognition.onerror = (event) => {
+            if (event.error === "not-allowed" || event.error === "service-not-allowed") {
+                setError({ code: event.error, message: "麦克风权限被拒绝，请在浏览器设置中允许访问麦克风" });
+            } else if (event.error === "no-speech") {
+                setError({ code: event.error, message: "未检测到语音，请重试" });
+            } else if (event.error === "network") {
+                setError({ code: event.error, message: "语音识别服务不可用，请检查网络后重试" });
+            } else {
+                setError({ code: event.error, message: `语音识别失败（${event.error}）` });
+            }
+        };
+        recognition.onend = () => {
+            if (stopTimerRef.current !== null) {
+                window.clearTimeout(stopTimerRef.current);
+                stopTimerRef.current = null;
+            }
+            const resolve = stopResolveRef.current;
+            stopResolveRef.current = null;
+            resolve?.(finalTextRef.current);
+        };
+        recognitionRef.current = recognition;
+        recognition.start();
+    }, [lang]);
+
+    const stop = useCallback((): Promise<string> => {
+        return new Promise((resolve) => {
+            const recognition = recognitionRef.current;
+            if (!recognition) {
+                resolve(finalTextRef.current);
+                return;
+            }
+            stopResolveRef.current = resolve;
+            // 兜底：部分浏览器 stop() 后 onend 可能不触发，超时后返回已识别文本
+            stopTimerRef.current = window.setTimeout(() => {
+                if (stopResolveRef.current) {
+                    stopResolveRef.current = null;
+                    resolve(finalTextRef.current);
+                }
+            }, 1500);
+            try {
+                recognition.stop();
+            } catch {
+                if (stopTimerRef.current !== null) {
+                    window.clearTimeout(stopTimerRef.current);
+                    stopTimerRef.current = null;
+                }
+                stopResolveRef.current = null;
+                resolve(finalTextRef.current);
+            }
+        });
+    }, []);
+
+    const cancel = useCallback(() => {
+        if (stopTimerRef.current !== null) {
+            window.clearTimeout(stopTimerRef.current);
+            stopTimerRef.current = null;
+        }
+        stopResolveRef.current = null;
+        const recognition = recognitionRef.current;
+        recognitionRef.current = null;
+        if (recognition) {
+            recognition.onend = null;
+            try {
+                recognition.abort();
+            } catch { /* ignore */ }
+        }
+        finalTextRef.current = "";
+    }, []);
+
+    useEffect(() => cancel, [cancel]);
+
+    return { supported, error, start, stop, cancel };
+}
diff --git a/web/src/hooks/use-voice-recording.ts b/web/src/hooks/use-voice-recording.ts
new file mode 100644
index 000000000..a8b58761f
--- /dev/null
+++ b/web/src/hooks/use-voice-recording.ts
@@ -0,0 +1,193 @@
+import { useCallback, useEffect, useRef, useState } from "react";
+
+export type VoiceRecordingState = "idle" | "recording" | "paused";
+
+type UseVoiceRecordingOptions = {
+    /** 最大录音时长（秒），默认 60 秒 */
+    maxDuration?: number;
+};
+
+export type UseVoiceRecordingReturn = {
+    state: VoiceRecordingState;
+    waveform: number[];
+    duration: number;
+    error: string | null;
+    start: () => Promise<void>;
+    stop: () => Promise<Blob | null>;
+    cancel: () => void;
+    reset: () => void;
+};
+
+/**
+ * 语音录制 Hook
+ * 使用 MediaRecorder API 录制音频，AnalyserNode 获取实时波形数据
+ */
+export function useVoiceRecording(options: UseVoiceRecordingOptions = {}): UseVoiceRecordingReturn {
+    const { maxDuration = 60 } = options;
+    const [state, setState] = useState<VoiceRecordingState>("idle");
+    const [waveform, setWaveform] = useState<number[]>([]);
+    const [duration, setDuration] = useState(0);
+    const [error, setError] = useState<string | null>(null);
+
+    const mediaRecorderRef = useRef<MediaRecorder | null>(null);
+    const audioContextRef = useRef<AudioContext | null>(null);
+    const analyserRef = useRef<AnalyserNode | null>(null);
+    const streamRef = useRef<MediaStream | null>(null);
+    const chunksRef = useRef<Blob[]>([]);
+    const animationFrameRef = useRef<number | null>(null);
+    const timerRef = useRef<number | null>(null);
+    const startTimeRef = useRef<number>(0);
+
+    // 清理资源
+    const cleanup = useCallback(() => {
+        if (animationFrameRef.current) {
+            cancelAnimationFrame(animationFrameRef.current);
+            animationFrameRef.current = null;
+        }
+        if (timerRef.current) {
+            window.clearInterval(timerRef.current);
+            timerRef.current = null;
+        }
+        if (streamRef.current) {
+            streamRef.current.getTracks().forEach((track) => track.stop());
+            streamRef.current = null;
+        }
+        if (audioContextRef.current) {
+            void audioContextRef.current.close();
+            audioContextRef.current = null;
+        }
+        analyserRef.current = null;
+        mediaRecorderRef.current = null;
+    }, []);
+
+    // 更新波形数据
+    const updateWaveform = useCallback(() => {
+        const analyser = analyserRef.current;
+        if (!analyser) return;
+
+        const bufferLength = analyser.frequencyBinCount;
+        const dataArray = new Uint8Array(bufferLength);
+        analyser.getByteFrequencyData(dataArray);
+
+        // 取 32 个频段的平均值，用于波形显示
+        const step = Math.floor(bufferLength / 32);
+        const values: number[] = [];
+        for (let i = 0; i < 32; i++) {
+            let sum = 0;
+            for (let j = 0; j < step; j++) {
+                sum += dataArray[i * step + j];
+            }
+            values.push(sum / step / 255);
+        }
+        setWaveform(values);
+
+        animationFrameRef.current = requestAnimationFrame(updateWaveform);
+    }, []);
+
+    // 开始录制
+    const start = useCallback(async () => {
+        setError(null);
+        setWaveform([]);
+        setDuration(0);
+        chunksRef.current = [];
+
+        try {
+            const stream = await navigator.mediaDevices.getUserMedia({ audio: true });
+            streamRef.current = stream;
+
+            const audioContext = new AudioContext();
+            audioContextRef.current = audioContext;
+
+            const source = audioContext.createMediaStreamSource(stream);
+            const analyser = audioContext.createAnalyser();
+            analyser.fftSize = 256;
+            source.connect(analyser);
+            analyserRef.current = analyser;
+
+            const mediaRecorder = new MediaRecorder(stream);
+            mediaRecorderRef.current = mediaRecorder;
+
+            mediaRecorder.ondataavailable = (event) => {
+                if (event.data.size > 0) {
+                    chunksRef.current.push(event.data);
+                }
+            };
+
+            mediaRecorder.start(100);
+            startTimeRef.current = Date.now();
+            setState("recording");
+
+            // 启动波形更新
+            animationFrameRef.current = requestAnimationFrame(updateWaveform);
+
+            // 启动计时器
+            timerRef.current = window.setInterval(() => {
+                const elapsed = (Date.now() - startTimeRef.current) / 1000;
+                setDuration(elapsed);
+                if (elapsed >= maxDuration) {
+                    // 达到最大时长，自动停止
+                    mediaRecorder.stop();
+                }
+            }, 100);
+        } catch (err) {
+            const message = err instanceof Error ? err.message : "无法访问麦克风";
+            setError(message);
+            setState("idle");
+            cleanup();
+        }
+
+    }, [cleanup, maxDuration, updateWaveform]);
+
+    // 停止录制
+    const stop = useCallback((): Promise<Blob | null> => {
+        return new Promise((resolve) => {
+            const mediaRecorder = mediaRecorderRef.current;
+            if (!mediaRecorder || mediaRecorder.state === "inactive") {
+                cleanup();
+                setState("idle");
+                resolve(null);
+                return;
+            }
+
+            mediaRecorder.onstop = () => {
+                const blob = chunksRef.current.length > 0 ? new Blob(chunksRef.current, { type: "audio/webm" }) : null;
+                cleanup();
+                setState("idle");
+                resolve(blob);
+            };
+
+            mediaRecorder.stop();
+        });
+    }, [cleanup]);
+
+    // 取消录制
+    const cancel = useCallback(() => {
+        chunksRef.current = [];
+        cleanup();
+        setState("idle");
+        setWaveform([]);
+        setDuration(0);
+    }, [cleanup]);
+
+    // 重置
+    const reset = useCallback(() => {
+        cancel();
+        setError(null);
+    }, [cancel]);
+
+    // 组件卸载时清理
+    useEffect(() => {
+        return () => cleanup();
+    }, [cleanup]);
+
+    return {
+        state,
+        waveform,
+        duration,
+        error,
+        start,
+        stop,
+        cancel,
+        reset,
+    };
+}
\ No newline at end of file
diff --git a/web/src/pages/create/index.tsx b/web/src/pages/create/index.tsx
index d96a37227..a0ae970b6 100644
--- a/web/src/pages/create/index.tsx
+++ b/web/src/pages/create/index.tsx
@@ -7,6 +7,7 @@ import { Link } from "react-router";
 
 import { AssetMediaPreview } from "@/components/asset-media-preview";
 import { CanvasResourceMentionTextarea } from "@/components/canvas/canvas-resource-mention-textarea";
+import { VoiceRecordingButton } from "@/components/conversation/voice-recording-button";
 import { ModelPicker } from "@/components/model-picker";
 import { canvasResourceMentionToken } from "@/lib/canvas/canvas-resource-references";
 import { createClientId } from "@/lib/client-id";
@@ -746,6 +747,10 @@ function CreationComposer(props: ComposerProps) {
         </div>
         <footer className="creation-chat-dock">
             <div className="creation-chat-controls">
+                <VoiceRecordingButton
+                    disabled={props.busy}
+                    onTranscribed={(text) => props.setPrompt(props.prompt.trim() ? `${props.prompt} ${text}` : text)}
+                />
                 <ModePicker mode={props.mode} onModeChange={props.onModeChange} />
                 <ModelPicker config={props.config} value={props.model} onChange={props.onModelChange} capability={props.mode} className="creation-model-picker" placeholder={`选择${modeLabels[props.mode]}模型`} showSelectedPrice={false} variant="creation" />
                 {props.mode === "video" || (props.mode === "image" && imageSettingsSupported) ? <GenerationSettingsMenu {...props} /> : null}
diff --git a/web/src/pages/test-voice-recording.tsx b/web/src/pages/test-voice-recording.tsx
new file mode 100644
index 000000000..a46d9f65f
--- /dev/null
+++ b/web/src/pages/test-voice-recording.tsx
@@ -0,0 +1,107 @@
+import { useState } from "react";
+import { Button, Card, message } from "antd";
+import { Mic, Send } from "lucide-react";
+
+import { VoiceRecordingButton } from "@/components/conversation/voice-recording-button";
+import { canvasThemes } from "@/lib/canvas-theme";
+import { useThemeStore } from "@/stores/use-theme-store";
+
+/**
+ * 语音录制功能测试页面
+ * 验证输入行内联波形录制和 STT 转写闭环
+ */
+export default function TestVoiceRecording() {
+    const theme = canvasThemes[useThemeStore((state) => state.theme)];
+    const [prompt, setPrompt] = useState("");
+    const [sending, setSending] = useState(false);
+
+    const handleTranscribed = (text: string) => {
+        // 转写结果填入输入框，供用户确认或直接发送
+        setPrompt((current) => (current.trim() ? `${current} ${text}` : text));
+        message.success("语音已转写为文字");
+    };
+
+    const handleSubmit = async () => {
+        if (!prompt.trim()) return;
+        setSending(true);
+        try {
+            // 原型阶段：发送到对话入口由调用方接入；此处仅输出
+            console.log("发送文本:", prompt);
+            message.success("发送成功（原型阶段，仅控制台输出）");
+            setPrompt("");
+        } finally {
+            setSending(false);
+        }
+    };
+
+
+    return (
+        <div className="min-h-screen p-8" style={{ background: theme.spatial.surface }}>
+            <div className="mx-auto max-w-2xl">
+                <Card
+                    title={
+                        <div className="flex items-center gap-2">
+                            <Mic className="size-5" style={{ color: theme.accent.primary }} />
+                            <span>实时对话功能测试（MVP）</span>
+                        </div>
+                    }
+                    style={{ background: theme.spatial.elevated, borderColor: theme.toolbar.border }}
+                >
+                    <div className="space-y-4">
+                        {/* 文本输入 */}
+                        <div>
+                            <label className="mb-2 block text-sm font-medium" style={{ color: theme.node.text }}>
+                                文本输入（语音转写结果会自动填入）
+                            </label>
+                            <textarea
+                                value={prompt}
+                                onChange={(e) => setPrompt(e.target.value)}
+                                placeholder="输入消息，或点击实时对话按钮用语音输入..."
+                                className="w-full rounded-lg border px-3 py-2 text-sm outline-none transition focus:border-current"
+                                style={{
+                                    background: theme.node.fill,
+                                    borderColor: theme.toolbar.border,
+                                    color: theme.node.text,
+                                    minHeight: 80,
+                                }}
+                            />
+                        </div>
+
+                        {/* 音频预览 */}
+
+                        {/* 控制栏：实时对话按钮（点击后在输入行展开波形录制条） */}
+                        <div className="flex items-center justify-between gap-2">
+                            <div className="flex flex-1 items-center gap-2">
+                                <VoiceRecordingButton onTranscribed={handleTranscribed} />
+                            </div>
+                            <Button
+                                type="primary"
+                                icon={<Send className="size-4" />}
+                                disabled={!prompt.trim()}
+                                loading={sending}
+                                onClick={handleSubmit}
+                                style={{ background: theme.accent.primary, borderColor: theme.accent.primary }}
+                            >
+                                发送
+                            </Button>
+                        </div>
+
+                        {/* 说明 */}
+                        <div className="rounded-lg border p-3 text-xs" style={{ background: theme.toolbar.panel, borderColor: theme.toolbar.border, color: theme.node.muted }}>
+                            <div className="font-semibold" style={{ color: theme.node.text }}>
+                                使用说明：
+                            </div>
+                            <ul className="mt-1 list-inside list-disc space-y-1">
+                                <li>点击麦克风按钮，输入行内展开波形录制条并自动开始录音</li>
+                                <li>波形动画实时显示音量变化</li>
+                                <li>点击停止按钮完成录制，自动上传并转写</li>
+                                <li>转写结果自动填入输入框，可编辑后发送</li>
+                                <li>转写失败时在录制条内提示，可点击麦克风重试</li>
+                            </ul>
+                        </div>
+                    </div>
+                </Card>
+            </div>
+        </div>
+    );
+}
\ No newline at end of file
diff --git a/web/src/router.tsx b/web/src/router.tsx
index 3a94c859b..00849c9be 100644
--- a/web/src/router.tsx
+++ b/web/src/router.tsx
@@ -30,6 +30,7 @@ import WalletPage from "@/pages/wallet";
 import ProjectsPage from "@/pages/projects";
 import ProjectDetailPage from "@/pages/projects/detail";
 import SettingsPage from "@/pages/settings";
+import TestVoiceRecording from "@/pages/test-voice-recording";
 
 export const router = createBrowserRouter([
     {
@@ -57,6 +58,7 @@ export const router = createBrowserRouter([
             { path: "/skills", element: <RequireAuth><SkillsPage /></RequireAuth> },
             { path: "/wallet", element: <RequireAuth><RequireFeature feature="creditsEnabled"><WalletPage /></RequireFeature></RequireAuth> },
             { path: "/settings", element: <RequireAuth><SettingsPage /></RequireAuth> },
+            { path: "/test-voice-recording", element: <RequireAuth><TestVoiceRecording /></RequireAuth> },
             { path: "/projects", element: <RequireAuth><RequireFeature feature="shortDramaEnabled"><ProjectsPage /></RequireFeature></RequireAuth> },
             { path: "/projects/:projectId", element: <RequireAuth><RequireFeature feature="shortDramaEnabled"><ProjectDetailPage /></RequireFeature></RequireAuth> },
             { path: "/projects/:projectId/:view", element: <RequireAuth><RequireFeature feature="shortDramaEnabled"><ProjectDetailPage /></RequireFeature></RequireAuth> },