dsh-audiogen 0.4.19 → 0.4.22

package/src/client/voices-view.tsx

@@ -12,10 +12,10 @@
 import { useEffect, useMemo, useState } from 'react'
 import type { AudiogenApi } from './api.ts'
 import type { AudiogenScope } from './settings-scope.ts'
-import type { VoiceEntry } from '../protocol.ts'
+import type { VoiceEntry, VoiceRecommendRecord } from '../protocol.ts'
 import { tt } from './helpers.ts'
 import { AudioPlayer } from './audio-player.tsx'
-import { MicIcon, SearchIcon, TrashIcon, CheckIcon } from './icons.tsx'
+import { MicIcon, SearchIcon, TrashIcon, CheckIcon, ListIcon } from './icons.tsx'
 import css from './voices.module.css'
 
 export interface VoicesReusePayload {
@@ -29,8 +29,26 @@ const SOURCE_LABELS: Record<VoiceEntry['source'], string> = {
   custom: '自建设计',
   owned: '账户自有',
   shared: '社区共享',
+  configured: '渠道模型',
 }
 
+/** 语言下拉：ISO 639-1 值；后端按两套厂商词汇匹配（ElevenLabs ISO / MiniMax 前缀）。 */
+const LANGUAGE_OPTIONS: Array<{ value: string; label: string }> = [
+  { value: '', label: '全部' },
+  { value: 'zh', label: '中文（普通话 / 粤语）' },
+  { value: 'en', label: '英语 English' },
+  { value: 'ja', label: '日语 Japanese' },
+  { value: 'ko', label: '韩语 Korean' },
+  { value: 'es', label: '西班牙语 Spanish' },
+  { value: 'fr', label: '法语 French' },
+  { value: 'de', label: '德语 German' },
+  { value: 'ru', label: '俄语 Russian' },
+  { value: 'it', label: '意大利语 Italian' },
+  { value: 'pt', label: '葡萄牙语 Portuguese' },
+  { value: 'ar', label: '阿拉伯语 Arabic' },
+  { value: 'hi', label: '印地语 Hindi' },
+]
+
 function useConfig(scope: AudiogenScope) {
   const [value, setValue] = useState(scope.getSnapshot().value)
   useEffect(() => scope.subscribe(() => { setValue(scope.getSnapshot().value) }), [scope])
@@ -82,6 +100,44 @@ export function VoicesView(props: {
   const [note, setNote] = useState<string | null>(null)
   const [deletingId, setDeletingId] = useState<string | null>(null)
 
+  // AI 推荐记录（每次推荐自动落盘，最多保留 50 条；关闭面板后仍可回看）
+  const [records, setRecords] = useState<VoiceRecommendRecord[]>([])
+  const [recordsOpen, setRecordsOpen] = useState(false)
+
+  const loadRecords = async (): Promise<void> => {
+    try {
+      const response = await api.voiceRecommendHistory(30)
+      setRecords(response.entries ?? [])
+    } catch {
+      // best-effort：记录加载失败不打扰主流程
+    }
+  }
+
+  const removeRecord = async (id: string): Promise<void> => {
+    await api.voiceRecommendHistoryRemove(id).catch(() => { /* best-effort */ })
+    void loadRecords()
+  }
+
+  useEffect(() => { void loadRecords() }, [api])
+
+  /** 当前的筛选载荷（查询音色列表用）。 */
+  const currentFilterPayload = (): Record<string, unknown> => ({
+    channel: channelId,
+    ...(language.trim() === '' ? {} : { language: language.trim() }),
+    ...(keyword.trim() === '' ? {} : { keyword: keyword.trim() }),
+    ...(source === '' ? {} : { source }),
+    ...(search.trim() === '' ? {} : { search: search.trim() }),
+    ...(useCase.trim() === '' ? {} : { use_case: useCase.trim() }),
+    ...(accent.trim() === '' ? {} : { accent: accent.trim() }),
+    ...(gender.trim() === '' ? {} : { gender: gender.trim() }),
+    ...(age.trim() === '' ? {} : { age: age.trim() }),
+    ...(locale.trim() === '' ? {} : { locale: locale.trim() }),
+    ...(category.trim() === '' ? {} : { category: category.trim() }),
+    ...(sort === '' ? {} : { sort }),
+    ...(featured ? { featured: true } : {}),
+    ...(freeUsersAllowed ? { free_users_allowed: true } : {}),
+  })
+
   const load = async (): Promise<void> => {
     const channel = channels.find(item => item.id === channelId)
     if (channel === undefined) return
@@ -90,20 +146,7 @@ export function VoicesView(props: {
     setNote(null)
     try {
       const response = await api.voiceList({
-        channel: channel.id,
-        ...(language.trim() === '' ? {} : { language: language.trim() }),
-        ...(keyword.trim() === '' ? {} : { keyword: keyword.trim() }),
-        ...(source === '' ? {} : { source }),
-        ...(search.trim() === '' ? {} : { search: search.trim() }),
-        ...(useCase.trim() === '' ? {} : { use_case: useCase.trim() }),
-        ...(accent.trim() === '' ? {} : { accent: accent.trim() }),
-        ...(gender.trim() === '' ? {} : { gender: gender.trim() }),
-        ...(age.trim() === '' ? {} : { age: age.trim() }),
-        ...(locale.trim() === '' ? {} : { locale: locale.trim() }),
-        ...(category.trim() === '' ? {} : { category: category.trim() }),
-        ...(sort === '' ? {} : { sort }),
-        ...(featured ? { featured: true } : {}),
-        ...(freeUsersAllowed ? { free_users_allowed: true } : {}),
+        ...currentFilterPayload(),
         limit: 500,
       })
       if (response.ok) {
@@ -158,6 +201,20 @@ export function VoicesView(props: {
     })
   }
 
+  /** 从推荐记录里复用音色：切到记录对应的渠道，再回填 TTS 表单。 */
+  const reuseFromRecord = (record: VoiceRecommendRecord, voice: { voice_id: string }): void => {
+    if (record.channel_id !== undefined && record.channel_id !== '' && channels.some(item => item.id === record.channel_id)) {
+      setChannelId(record.channel_id)
+    }
+    const channel = channels.find(item => item.id === (record.channel_id ?? channelId))
+    const ttsModel = channel?.models.find(model => model.category === 'tts' || /tts|speech|voice|t2a/i.test(model.alias))
+    onReuseVoice({
+      mode: 'tts',
+      voiceId: voice.voice_id,
+      ...(ttsModel === undefined || ttsModel.alias === '' ? {} : { model: ttsModel.alias }),
+    })
+  }
+
   return (
     <div className={css.view}>
       <div className={css.toolbar}>
@@ -171,7 +228,11 @@ export function VoicesView(props: {
         </label>
         <label className={css.field}>
           <span className={css.fieldLabel}>语言</span>
-          <input className={css.input} value={language} onChange={event => setLanguage(event.target.value)} placeholder="en / zh / ja…" />
+          <select className={css.select} value={language} onChange={event => setLanguage(event.target.value)}>
+            {LANGUAGE_OPTIONS.map(option => (
+              <option key={option.value} value={option.value}>{option.label}</option>
+            ))}
+          </select>
         </label>
         <label className={css.field}>
           <span className={css.fieldLabel}>关键词</span>
@@ -192,6 +253,57 @@ export function VoicesView(props: {
         </button>
       </div>
 
+      <p className={css.stateNote}>
+        按需求描述（prompt）推荐音色请使用生成页左侧「🎤 音色推荐」模式；这里仅浏览/筛选/删除音色，下方为历次 AI 推荐记录。
+      </p>
+
+      {records.length > 0 ? (
+        <details
+          className={css.recordBox}
+          open={recordsOpen}
+          onToggle={(event) => setRecordsOpen((event.target as HTMLDetailsElement).open)}
+        >
+          <summary className={css.recordSummary}>
+            <ListIcon /> 最近 AI 推荐（{records.length}）
+            <span className={css.recordHint}>每次推荐自动记录，点「用此音色生成」可直接回填 TTS 表单</span>
+          </summary>
+          <div className={css.recordList}>
+            {records.map(record => (
+              <div className={`${css.card} ${css.recordCard}`} key={record.id}>
+                <div className={css.cardHead}>
+                  <span className={css.voiceName} title={record.requirement}>{record.requirement}</span>
+                  <span className={css.badge} data-source={record.vendor}>{record.channel}</span>
+                </div>
+                <div className={css.meta}>
+                  <span>{new Date(record.createdAt).toLocaleString()}</span>
+                  <span>候选池 {record.candidate_count}</span>
+                  <span>推荐 {record.recommendations.length} 条</span>
+                </div>
+                <div className={css.recordVoices}>
+                  {record.recommendations.map((voice, index) => (
+                    <div className={css.recordVoiceRow} key={`${record.id}:${voice.voice_id}`}>
+                      <span className={css.rank}>{index + 1}</span>
+                      <span className={css.recordVoiceName} title={voice.name}>{voice.name}</span>
+                      <span className={css.voiceId} title={voice.voice_id}>ID: {voice.voice_id}</span>
+                      {voice.descriptive === undefined ? null : <span className={css.reasonInline}>descriptive: {voice.descriptive}</span>}
+                      {voice.reason === '' ? null : <span className={css.reasonInline}>{voice.reason}</span>}
+                      <button type="button" className={css.useBtn} onClick={() => reuseFromRecord(record, voice)}>
+                        <MicIcon /> 用此音色生成
+                      </button>
+                    </div>
+                  ))}
+                </div>
+                <div className={css.actions}>
+                  <button type="button" className={css.deleteBtn} onClick={() => void removeRecord(record.id)}>
+                    <TrashIcon /> 删除记录
+                  </button>
+                </div>
+              </div>
+            ))}
+          </div>
+        </details>
+      ) : null}
+
       <details className={css.advanced}>
         <summary className={css.advancedSummary}>官方共享库筛选（ElevenLabs /v1/shared-voices，其它渠道无效）</summary>
         <div className={css.advancedGrid}>

package/src/client/voices.module.css

@@ -249,3 +249,168 @@
   font-size: 12.5px;
   color: var(--dsw-alias-fg-l3, #6b7280);
 }
+
+/* AI 推荐（按需求描述选音色） */
+
+.recommend {
+  border: 1px solid var(--dsw-alias-border-l2, #e5e7eb);
+  border-radius: 10px;
+  padding: 10px 12px;
+  background: var(--dsw-alias-bg-layer-2, #f9fafb);
+  display: flex;
+  flex-direction: column;
+  gap: 8px;
+}
+
+.recommendHead {
+  display: flex;
+  align-items: center;
+  gap: 6px;
+  flex-wrap: wrap;
+  color: var(--dsw-accent, #2563eb);
+}
+
+.recommendTitle {
+  font-size: 13px;
+  font-weight: 600;
+  color: var(--dsw-alias-fg, #111827);
+}
+
+.recommendHint {
+  font-size: 11.5px;
+  color: var(--dsw-alias-fg-l3, #6b7280);
+}
+
+.recommendBody {
+  display: flex;
+  flex-direction: column;
+  gap: 8px;
+}
+
+.recommendInput {
+  width: 100%;
+  min-height: 48px;
+  padding: 7px 9px;
+  border: 1px solid var(--dsw-alias-border-l2, #d1d5db);
+  border-radius: 8px;
+  background: var(--dsw-alias-bg, #fff);
+  color: var(--dsw-alias-fg, #111827);
+  font-size: 13px;
+  font-family: inherit;
+  resize: vertical;
+  outline: none;
+}
+
+.recommendInput:focus {
+  border-color: var(--dsw-accent, #2563eb);
+}
+
+.recommendControls {
+  display: flex;
+  align-items: end;
+  gap: 8px;
+}
+
+.recommendResults {
+  display: flex;
+  flex-direction: column;
+  gap: 8px;
+  margin-top: 2px;
+}
+
+.recommendCard {
+  border-color: var(--dsw-accent, #2563eb);
+  background: color-mix(in srgb, var(--dsw-accent, #2563eb) 4%, var(--dsw-alias-bg, #fff));
+}
+
+.rank {
+  display: inline-flex;
+  align-items: center;
+  justify-content: center;
+  width: 20px;
+  height: 20px;
+  border-radius: 999px;
+  background: var(--dsw-accent, #2563eb);
+  color: #fff;
+  font-size: 11px;
+  font-weight: 600;
+  flex: none;
+}
+
+.reason {
+  margin: 0;
+  font-size: 12.5px;
+  line-height: 1.5;
+  color: var(--dsw-accent, #2563eb);
+}
+
+/* ------------------------------------------------------ AI 推荐记录 */
+.recordBox {
+  border: 1px solid var(--dsw-alias-border-l2, #e5e7eb);
+  border-radius: 10px;
+  background: var(--dsw-alias-bg-layer-2, #f9fafb);
+  padding: 8px 12px;
+}
+
+.recordSummary {
+  display: flex;
+  align-items: center;
+  gap: 6px;
+  cursor: pointer;
+  font-size: 13px;
+  font-weight: 600;
+  color: var(--dsw-alias-fg, #111827);
+  user-select: none;
+}
+
+.recordHint {
+  font-size: 11.5px;
+  font-weight: 400;
+  color: var(--dsw-alias-fg-l3, #6b7280);
+}
+
+.recordList {
+  display: flex;
+  flex-direction: column;
+  gap: 8px;
+  margin-top: 8px;
+}
+
+.recordCard {
+  border-left: 3px solid var(--dsw-accent, #2563eb);
+}
+
+.recordVoices {
+  display: flex;
+  flex-direction: column;
+  gap: 4px;
+  margin-top: 6px;
+}
+
+.recordVoiceRow {
+  display: flex;
+  align-items: center;
+  gap: 8px;
+  flex-wrap: wrap;
+  padding: 4px 2px;
+  border-radius: 6px;
+}
+
+.recordVoiceRow:hover {
+  background: color-mix(in srgb, var(--dsw-accent, #2563eb) 5%, transparent);
+}
+
+.recordVoiceName {
+  font-size: 13px;
+  font-weight: 500;
+  color: var(--dsw-alias-fg, #111827);
+  overflow: hidden;
+  text-overflow: ellipsis;
+  white-space: nowrap;
+  max-width: 220px;
+}
+
+.reasonInline {
+  font-size: 12px;
+  color: var(--dsw-accent, #2563eb);
+}

package/src/index.ts

@@ -22,6 +22,8 @@ import { enhancePromptText } from './prompt-enhance.ts'
 import { AudioGenError } from './audio-engine.ts'
 import { makeRoutes, type ChannelsView, type SettingsSeam } from './routes.ts'
 import type { AudioChannel } from './audio-engine.ts'
+import { recommendVoices, type VoiceRecommendation } from './voice-recommend.ts'
+import type { VendorVoiceEntry } from './voice-manager.ts'
 import { registerAgentAudioTools, type AgentAudioToolConfig } from './agent-audio-tools.ts'
 import { audioPresetById } from './audio-presets.ts'
 
@@ -82,7 +84,7 @@ const DEFAULT_ALLOW_AGENT_AUDIO = true
 
 const SECTION_ORDER = 160
 
-export const AUDIOGEN_GUIDANCE = '本机已安装 dsh-audiogen 插件（DSH AI 音频）：侧边栏「AI 音频」入口。能力：通过「渠道」对接多个音频生成厂商（OpenAI TTS、ElevenLabs、MiniMax、Stability Audio、自定义 OpenAI 兼容接口），支持 TTS 文本转语音、音乐生成和音效生成。API 地址与密钥在 GUI 设置中按渠道配置，密钥仅存于本机设置文档；生成请求由本地宿主代理转发。Agent 可直接调用 `generate_audio` 提交 TTS/音乐/音效任务，默认等待完成并返回同源音频 URL；可用 `manage_audio_voices` 浏览/筛选/删除厂商音色（MiniMax、ElevenLabs），再用选定音色的 voice_id 调用 `generate_audio` 生成。限制：生成消耗上游 API 额度；音频内容由上游模型生成；模型只能使用用户在各渠道配置目录中的模型。用户提到「音频 / 语音 / TTS / 配乐 / 音效 / AI 音频」时即指本插件，请据此协作。'
+export const AUDIOGEN_GUIDANCE = '本机已安装 dsh-audiogen 插件（DSH AI 音频）：侧边栏「AI 音频」入口。能力：通过「渠道」对接多个音频生成厂商（OpenAI TTS、ElevenLabs、MiniMax、Stability Audio、自定义 OpenAI 兼容接口），支持 TTS 文本转语音、音乐生成和音效生成。API 地址与密钥在 GUI 设置中按渠道配置，密钥仅存于本机设置文档；生成请求由本地宿主代理转发。Agent 可直接调用 `generate_audio` 提交 TTS/音乐/音效任务，默认等待完成并返回同源音频 URL；可用 `manage_audio_voices` 浏览/筛选/删除厂商音色（MiniMax、ElevenLabs），也能用该工具的 action=recommend 按自然语言需求描述（如「清亮甜美的少女音」）让默认模型推荐 top-k 音色，再用选定音色的 voice_id 调用 `generate_audio` 生成。角色音色选角（小说/游戏配音 casting）：传角色画像（JSON 数组/对象，字段 character_id/character_name/gender/age_stage/voice_traits/personality_traits/appearance/sample_lines/dialogue_count/language；文本描述需先整理成该结构）调用 `manage_audio_voices` 的 action=cast，工具按性别/年龄/用途做确定性硬过滤（accent 为偏好、候选为空才放松）并返回每个角色的候选池，你在上下文中全局权衡主音色（lead/major 不复用）+ 备用音色后，用 action=save_cast 校验落盘（~/.dsh/dsh-audiogen/cast-selections.json），随后按选定 voice_id 生成 TTS；MiniMax 也可先用 `generate_audio(mode=voice_design)` 为角色创作专属音色再选角。限制：生成消耗上游 API 额度；音频内容由上游模型生成；模型只能使用用户在各渠道配置目录中的模型；音色管理仅 MiniMax 与 ElevenLabs 渠道支持。用户提到「音频 / 语音 / TTS / 配乐 / 音效 / AI 音频」时即指本插件，请据此协作。'
 
 function guidanceFor(channels: AudioChannel[], defaultChannelId: string): string {
   if (channels.length === 0) {
@@ -225,6 +227,17 @@ export function apply(ctx: Context, config?: Config): void {
     return provider !== '' && model !== '' ? { provider, model } : undefined
   }
 
+  // 音色推荐：与提示词增强同机制，复用 Agent 默认模型（agent-default-model）。
+  const recommend = async (
+    requirement: string,
+    candidates: VendorVoiceEntry[],
+    topK: number,
+  ): Promise<VoiceRecommendation[]> => {
+    const seam = ctx.get('settings') as unknown as SettingsSeam
+    if (seam?.describe === undefined) throw new AudioGenError('设置服务不可用，无法推荐音色', 'settings-unavailable')
+    return recommendVoices({ settings: seam, llm: () => ctx.get('llm') }, requirement, candidates, topK)
+  }
+
   /** 读取「设置 → 模型」目录：各提供方 + 可广播的模型列表（增强模型下拉候选）。 */
   const llmModelOptions = async (): Promise<LlmModelOption[]> => {
     const llm = ctx.get('llm') as {
@@ -271,6 +284,7 @@ export function apply(ctx: Context, config?: Config): void {
         autoSave: () => resolve().autoSaveToLibrary,
         budget,
         enhance,
+        recommend,
         llmModelOptions,
       })
       const disposers = routes.map(route => ctx.webServer.register(route))
@@ -289,6 +303,7 @@ export function apply(ctx: Context, config?: Config): void {
         autoSaveToLibrary: value.autoSaveToLibrary,
         budget,
         enhance,
+        recommend,
       }
     }), 'dsh-audiogen: agent audio tools')
   })

package/src/protocol.ts

@@ -8,7 +8,7 @@
 export const AUDIOGEN_SETTINGS_NAMESPACE = 'dsh-audiogen'
 
 /** Published package version shared by the host updater and the client UI. */
-export const PLUGIN_VERSION = '0.4.19'
+export const PLUGIN_VERSION = '0.4.22'
 
 /** Same-origin route family (loopback-only, mirroring dsh-imagegen). */
 export const SETTINGS_API = {
@@ -42,15 +42,57 @@ export const MODEL_API = {
 export const VOICES_API = {
   list: '/api/dsh-audiogen/voices/list',
   delete: '/api/dsh-audiogen/voices/delete',
+  recommend: '/api/dsh-audiogen/voices/recommend',
+  recommendHistory: {
+    list: '/api/dsh-audiogen/voices/recommend-history/list',
+    remove: '/api/dsh-audiogen/voices/recommend-history/remove',
+  },
 } as const
 
+/** 一条已记录的 AI 音色推荐结果（面板与 Agent 共用同一存储）。 */
+export interface VoiceRecommendRecord {
+  id: string
+  createdAt: number
+  /** 渠道展示名（如 MiniMax / ElevenLabs）。 */
+  channel: string
+  /** 渠道 id（「用此音色生成」时切换目标渠道用）。 */
+  channel_id?: string
+  vendor: string
+  requirement: string
+  /** 候选池大小（推荐时）。 */
+  candidate_count: number
+  top_k: number
+  /** 推荐时使用的筛选条件（language/keyword/use_case/accent… 快照）。 */
+  filters?: Record<string, string | boolean>
+  recommendations: Array<{
+    voice_id: string
+    name: string
+    source: string
+    deletable: boolean
+    language?: string
+    locale?: string
+    accent?: string
+    gender?: string
+    age?: string
+    use_case?: string
+    category?: string
+    /** 厂商返回的完整标签（ElevenLabs shared-voices labels 原样保留）。 */
+    labels?: Record<string, string>
+    /** ElevenLabs descriptive 标签（如 serious / calm / whispery），选角排序参考。 */
+    descriptive?: string
+    description?: string
+    preview_url?: string
+    reason: string
+  }>
+}
+
 /** One normalized vendor voice entry (see voice-manager). */
 export interface VoiceEntry {
   provider: string
   voice_id: string
   name: string
-  /** system | custom | owned | shared — determines deletability. */
-  source: 'system' | 'custom' | 'owned' | 'shared'
+  /** system | custom | owned | shared | configured — determines deletability. */
+  source: 'system' | 'custom' | 'owned' | 'shared' | 'configured'
   deletable: boolean
   language?: string
   locale?: string
@@ -59,6 +101,10 @@ export interface VoiceEntry {
   age?: string
   use_case?: string
   category?: string
+  /** 厂商返回的完整标签（ElevenLabs labels 原样保留）。 */
+  labels?: Record<string, string>
+  /** ElevenLabs descriptive 标签（如 serious / calm / whispery）。 */
+  descriptive?: string
   description?: string
   preview_url?: string
 }
@@ -93,6 +139,19 @@ export interface VoicesDeleteRequest {
   confirm: boolean
 }
 
+/** Prompt-based recommendation request: 需求描述 + 可选的候选池缩小条件。 */
+export interface VoicesRecommendRequest extends VoicesListRequest {
+  /** 自然语言需求描述（如「17岁清亮甜美的少女音，适合活泼女主角」）。 */
+  requirement: string
+  /** 推荐条数（1-10，默认 5）。 */
+  top_k?: number
+}
+
+/** 单条推荐：音色条目 + LLM 理由（复用 VoiceEntry 形状以便前端直接渲染/复用）。 */
+export interface VoiceRecommendation extends VoiceEntry {
+  reason: string
+}
+
 /** Loopback-only audio file reader for panel/tool-result previews. */
 export const AUDIO_API = {
   file: '/api/dsh-audiogen/audio',