feat: add microphone input selection

Author: yexia553

electron/main/README.md

@@ -1,5 +1,12 @@
 # main/
 
+## Microphone Selection
+
+The selected microphone is stored by `config-manager.ts` as part of ASR config.
+`audio/session-manager.ts` reads that value when a recording session starts and
+forwards the device ID to the hidden recording renderer. An empty value keeps the
+existing system-default microphone behavior.
+
 Electron 主进程目录，负责窗口管理、IPC、录音编排、ASR/润色调用与文本注入。
 
 ## 文件列表

electron/main/audio/README.md

@@ -5,14 +5,15 @@ Main-process audio pipeline for recording sessions and chunked transcription.
 ## Files
 
 - `index.ts` - Re-exports the audio module surface.
-- `session-manager.ts` - Owns the active recording session lifecycle, `sessionId`, and HUD state transitions, including the initial transcribing step after recording stops. Runs a 10s watchdog after SESSION_STOP that aborts the session if the final chunk never arrives, and exposes `handleBackgroundRendererGone` to fail the active session when the hidden recording renderer crashes or hangs.
+- `session-manager.ts` - Owns the active recording session lifecycle, `sessionId`, selected microphone device forwarding, and HUD state transitions, including the initial transcribing step after recording stops. Runs a 10s watchdog after SESSION_STOP that aborts the session if the final chunk never arrives, and exposes `handleBackgroundRendererGone` to fail the active session when the hidden recording renderer crashes or hangs.
 - `processor.ts` - Accepts audio chunks, writes temp files, converts to MP3 for GLM or 16k mono WAV for local SenseVoice, calls the configured ASR provider, merges chunk text in order, promotes the HUD into the refine step when applicable, logs final line-break metadata, and runs the final refine/history/inject step once. Empty chunk buffers are recorded as empty transcripts (renderer placeholder markers); an empty final transcript completes the session without refine/history/injection. Exposes `hasReceivedFinalChunk`/`abortChunkSession` for the session watchdog.
 - `converter.ts` - Initializes FFmpeg and converts captured audio to MP3 or 16k mono WAV, with optional low-volume gain.
 
 ## Current Flow
 
-1. The renderer records one session for up to 3 minutes and rotates internal chunks every 29 seconds.
-2. The main process tracks chunk work by `sessionId + chunkIndex` and can process chunk ASR requests out of order.
-3. Finalization only runs after the final chunk has been seen and every chunk from `0..finalChunkIndex` has produced text.
-4. Refinement, line-break-aware final text logging, history writes, and text injection happen once per session after the merged transcript is ready.
-5. Any chunk failure or session cancellation aborts the session and discards late results.
+1. The main process starts a session and sends the hidden recorder the `sessionId` plus the saved microphone device ID when one is selected.
+2. The renderer records one session for up to 3 minutes and rotates internal chunks every 29 seconds.
+3. The main process tracks chunk work by `sessionId + chunkIndex` and can process chunk ASR requests out of order.
+4. Finalization only runs after the final chunk has been seen and every chunk from `0..finalChunkIndex` has produced text.
+5. Refinement, line-break-aware final text logging, history writes, and text injection happen once per session after the merged transcript is ready.
+6. Any chunk failure or session cancellation aborts the session and discards late results.

electron/main/audio/session-manager.ts

@@ -2,6 +2,7 @@ import { IPC_CHANNELS, type RecordingStartPayload, type VoiceSession } from '../
 import { showOverlay, hideOverlay, updateOverlay, showErrorAndHide } from '../window/overlay'
 import { getBackgroundWindow } from '../window/background'
 import { t } from '../i18n'
+import { configManager } from '../config-manager'
 import { abortChunkSession, hasReceivedFinalChunk } from './processor'
 
 // How long after SESSION_STOP the final audio chunk may take to arrive before
@@ -102,7 +103,11 @@ export async function handleStartRecording(): Promise<void> {
       return
     }
 
-    const payload: RecordingStartPayload = { sessionId: currentSession.id }
+    const asrConfig = configManager.getASRConfig()
+    const payload: RecordingStartPayload = {
+      sessionId: currentSession.id,
+      microphoneDeviceId: asrConfig.microphoneDeviceId || undefined,
+    }
     bgWindow.webContents.send(IPC_CHANNELS.SESSION_START, payload)
     const duration = Date.now() - startTimestamp
     console.log(`[Audio:Session] Recording start completed in ${duration}ms`)

electron/main/config-manager.ts

@@ -10,7 +10,7 @@ import {
   TranslationConfig,
 } from '../shared/types'
 import { normalizeRefineBaseUrl } from '../shared/refine-url'
-import { DEFAULT_HOTKEYS, LLM_REFINE, TRANSLATION } from '../shared/constants'
+import { DEFAULT_HOTKEYS, LLM_REFINE, MICROPHONE_INPUT, TRANSLATION } from '../shared/constants'
 
 const ENCRYPTED_PREFIX = 'enc:'
 
@@ -69,6 +69,8 @@ const defaultConfig: AppConfig = {
       intl: '',
     },
     lowVolumeMode: true,
+    microphoneDeviceId: '',
+    microphoneDeviceLabel: '',
     endpoint: '',
     language: 'auto',
   },
@@ -156,6 +158,10 @@ function normalizeASRProvider(provider: unknown): ASRProviderType {
   return provider === 'local-sensevoice' ? 'local-sensevoice' : 'glm'
 }
 
+function normalizeConfigString(value: unknown, maxLength: number): string {
+  return typeof value === 'string' ? value.trim().slice(0, maxLength) : ''
+}
+
 export class ConfigManager {
   private store: Store<ConfigSchema>
 
@@ -275,6 +281,14 @@ export class ConfigManager {
         ...(storedConfig.apiKeys ?? {}),
       },
     }
+    config.microphoneDeviceId = normalizeConfigString(
+      config.microphoneDeviceId,
+      MICROPHONE_INPUT.DEVICE_ID_MAX_LENGTH,
+    )
+    config.microphoneDeviceLabel = normalizeConfigString(
+      config.microphoneDeviceLabel,
+      MICROPHONE_INPUT.DEVICE_LABEL_MAX_LENGTH,
+    )
     config.apiKeys = {
       cn: this.decryptKey(config.apiKeys.cn),
       intl: this.decryptKey(config.apiKeys.intl),
@@ -294,6 +308,14 @@ export class ConfigManager {
       ...current,
       ...config,
       provider: normalizeASRProvider(config.provider ?? current.provider),
+      microphoneDeviceId: normalizeConfigString(
+        config.microphoneDeviceId ?? current.microphoneDeviceId,
+        MICROPHONE_INPUT.DEVICE_ID_MAX_LENGTH,
+      ),
+      microphoneDeviceLabel: normalizeConfigString(
+        config.microphoneDeviceLabel ?? current.microphoneDeviceLabel,
+        MICROPHONE_INPUT.DEVICE_LABEL_MAX_LENGTH,
+      ),
       apiKeys: {
         ...current.apiKeys,
         ...(config.apiKeys ?? {}),

electron/preload/README.md

@@ -1,5 +1,10 @@
 # preload/
 
+## Microphone Selection
+
+`RecordingStartPayload` may include `microphoneDeviceId`; the hidden recorder
+uses it as a capture hint while preload keeps the same listener API.
+
 Electron 预加载脚本目录，作为主进程与渲染进程之间的安全桥梁。
 
 ## 文件列表

electron/shared/README.md

@@ -1,5 +1,12 @@
 # shared/
 
+## Microphone Selection
+
+`ASRConfig` stores the selected microphone device ID and last known label.
+`RecordingStartPayload` carries the device ID from main to the hidden recorder,
+and `MICROPHONE_INPUT` defines the renderer select sentinel plus config string
+length caps.
+
 主进程与渲染进程共享的类型、常量与多语言资源。
 
 ## 文件列表

electron/shared/constants.ts

@@ -320,6 +320,12 @@ export const AUDIO_CONFIG = {
   BIT_DEPTH: 16,
 } as const
 
+export const MICROPHONE_INPUT = {
+  SYSTEM_DEFAULT_ID: '__system-default__',
+  DEVICE_ID_MAX_LENGTH: 256,
+  DEVICE_LABEL_MAX_LENGTH: 128,
+} as const
+
 export const LOW_VOLUME_GAIN_DB = 10
 
 export const HISTORY_RETENTION_DAYS = 90

electron/shared/locales/en.json

@@ -142,6 +142,22 @@
     "durationWarning": "Note: Maximum duration for single voice input is 3 minutes.",
     "lowVolumeMode": "Low-volume mode",
     "lowVolumeModeHelp": "Boost quiet speech in software before ASR (may also amplify background noise).",
+    "microphone": {
+      "label": "Microphone",
+      "placeholder": "Select microphone",
+      "systemDefault": "System default microphone",
+      "refresh": "Detect",
+      "detecting": "Detecting...",
+      "help": "Choose which input device Voice Key records from.",
+      "noDevices": "No microphones detected",
+      "deviceFallback": "Microphone {{index}}",
+      "unknownDevice": "Selected microphone",
+      "unavailableOption": "{{label}} (unavailable)",
+      "unavailableHelp": "The selected microphone is unavailable. Recording will fall back to the system default until it returns or you choose another device.",
+      "permissionHelp": "Microphone permission is needed to show device names.",
+      "unsupported": "This environment cannot detect microphone devices.",
+      "detectFailed": "Couldn't detect microphones."
+    },
     "asrProvider": "Recognition engine",
     "asrProviderLocal": "Local model",
     "asrProviderGlm": "Zhipu GLM",

electron/shared/locales/zh.json

@@ -142,6 +142,22 @@
     "durationWarning": "注意：单次语音输入最长为 3 分钟。",
     "lowVolumeMode": "低音量模式",
     "lowVolumeModeHelp": "在调用 ASR 前对小声语音进行软件增益（可能同时放大环境噪声）。",
+    "microphone": {
+      "label": "麦克风",
+      "placeholder": "选择麦克风",
+      "systemDefault": "系统默认麦克风",
+      "refresh": "检测",
+      "detecting": "检测中...",
+      "help": "选择 Voice Key 录音时使用的输入设备。",
+      "noDevices": "未检测到麦克风",
+      "deviceFallback": "麦克风 {{index}}",
+      "unknownDevice": "已选择的麦克风",
+      "unavailableOption": "{{label}}（不可用）",
+      "unavailableHelp": "已选择的麦克风当前不可用。录音会暂时回退到系统默认麦克风，直到设备恢复或你选择其它设备。",
+      "permissionHelp": "需要麦克风权限才能显示设备名称。",
+      "unsupported": "当前环境无法检测麦克风设备。",
+      "detectFailed": "无法检测麦克风。"
+    },
     "asrProvider": "识别引擎",
     "asrProviderLocal": "本地模型",
     "asrProviderGlm": "智谱 GLM",

electron/shared/types.ts

@@ -20,6 +20,8 @@ export interface ASRConfig {
     intl: string
   }
   lowVolumeMode?: boolean
+  microphoneDeviceId?: string
+  microphoneDeviceLabel?: string
 
   // Deprecated: for backward compatibility during migration
   apiKey?: string
@@ -125,6 +127,7 @@ export interface RefineConnectionResult {
 
 export interface RecordingStartPayload {
   sessionId: string
+  microphoneDeviceId?: string
 }
 
 export interface AudioChunkPayload {