dsh-audiogen 0.1.0 → 0.3.0

package/src/client/SettingsCard.tsx

@@ -12,7 +12,7 @@ import { createSnapshotStore, type SnapshotStore } from '@deepseek-ai/dsh-client
 import { CardForm, booleanField, textField, type CardActions, type CardShell, type FieldState as CardFieldState } from './settings-form.ts'
 import { ChannelsForm, type ChannelDraft, type ChannelsFormActions, type ChannelsFormState } from './channels-form.ts'
 import type { AudiogenScope } from './settings-scope.ts'
-import { PRESETS_API, type ModelMapping, type PresetProviderView } from '../protocol.ts'
+import { MODEL_API, PRESETS_API, type DiscoveredAudioModel, type ModelMapping, type PresetProviderView } from '../protocol.ts'
 import type { AudioGenKey } from './locales.ts'
 import css from './settings-card.module.css'
 
@@ -98,14 +98,22 @@ function newChannelDraft(preset: PresetProviderView | undefined, existing: Chann
 }
 
 function modelsToText(models: ModelMapping[]): string {
-  return models.map(model => `${model.alias}=${model.id}`).join('\n')
+  return models.map(model => `${model.alias}=${model.id}${model.category === undefined ? '' : ` @${model.category}`}`).join('\n')
 }
 
 function textToModels(text: string): ModelMapping[] {
   return text.split(/\n|,/).map(line => line.trim()).filter(Boolean).map(line => {
-    const eq = line.indexOf('=')
-    if (eq < 0) return { alias: line, id: line }
-    return { alias: line.slice(0, eq).trim(), id: line.slice(eq + 1).trim() }
+    const at = line.lastIndexOf(' @')
+    const category = at >= 0 ? line.slice(at + 2).trim() : undefined
+    const body = at >= 0 ? line.slice(0, at).trim() : line
+    const eq = body.indexOf('=')
+    const alias = eq >= 0 ? body.slice(0, eq).trim() : body.trim()
+    const id = eq >= 0 ? body.slice(eq + 1).trim() : alias
+    return {
+      alias,
+      id: id === '' ? alias : id,
+      ...(category === undefined || category === '' ? {} : { category: category as NonNullable<ModelMapping['category']> }),
+    }
   }).filter(model => model.alias !== '')
 }
 
@@ -118,6 +126,8 @@ export function AudioGenSettingsCard(props: AudioGenSettingsCardProps) {
   const [presets, setPresets] = useState<PresetProviderView[]>([])
   const [presetError, setPresetError] = useState<string | null>(null)
   const [confirmDeleteId, setConfirmDeleteId] = useState<string | null>(null)
+  const [discovering, setDiscovering] = useState(false)
+  const [discoverError, setDiscoverError] = useState<string | null>(null)
 
   const [editName, setEditName] = useState('')
   const [editUrl, setEditUrl] = useState('')
@@ -161,6 +171,37 @@ export function AudioGenSettingsCard(props: AudioGenSettingsCardProps) {
     setEditingId(null)
   }
 
+  const discoverModels = async (): Promise<void> => {
+    if (editingId === null) return
+    setDiscovering(true)
+    setDiscoverError(null)
+    try {
+      const existing = channels.find(channel => channel.id === editingId)
+      const response = await fetch(MODEL_API.discover, {
+        method: 'POST',
+        headers: { 'content-type': 'application/json' },
+        body: JSON.stringify({
+          channelId: editingId,
+          ...(editUrl.trim() !== '' ? { apiUrl: editUrl.trim() } : {}),
+          ...(editKey.trim() !== '' ? { apiKey: editKey.trim() } : {}),
+        }),
+      })
+      const body = await response.json() as { ok?: boolean; models?: DiscoveredAudioModel[]; message?: string; source?: string }
+      if (body.ok !== true || body.models === undefined) {
+        throw new Error(body.message ?? `HTTP ${response.status}`)
+      }
+      setEditModels(modelsToText([
+        ...body.models,
+        ...textToModels(editModels).filter(existingModel => !body.models!.some(model => model.id === existingModel.id)),
+      ]))
+      void existing
+    } catch (error) {
+      setDiscoverError(error instanceof Error ? error.message : String(error))
+    } finally {
+      setDiscovering(false)
+    }
+  }
+
   return (
     <li className={css.card}>
       <button
@@ -297,6 +338,12 @@ export function AudioGenSettingsCard(props: AudioGenSettingsCardProps) {
                 <label className={css.label} htmlFor={`audiogen-models-${editing.id}`}>{t('channel.models')}</label>
                 <textarea id={`audiogen-models-${editing.id}`} className={css.textarea} value={editModels} onChange={event => setEditModels(event.target.value)} />
                 <p className={css.sectionHint}>{t('channel.modelsHint')}</p>
+                <div className={css.channelAddRow}>
+                  <button type="button" className={css.channelAdd} disabled={discovering || !state.writable} onClick={() => void discoverModels()}>
+                    {discovering ? '获取中…' : '获取可用模型'}
+                  </button>
+                </div>
+                {discoverError !== null ? <p className={css.failed}>{discoverError}</p> : null}
               </div>
               <label className={css.label}>
                 <input type="checkbox" checked={editDefault} onChange={event => setEditDefault(event.target.checked)} /> {t('channel.default')}

package/src/client/channels-form.ts

@@ -78,7 +78,11 @@ function deepEqualJson(a: unknown, b: unknown): boolean {
 /** Trim and normalize a draft channel (models never carry empty aliases). */
 function stripChannel(channel: ChannelDraft): ChannelDraft {
   const models = channel.models
-    .map(model => ({ alias: model.alias.trim(), id: model.id.trim() === '' ? model.alias.trim() : model.id.trim() }))
+    .map(model => ({
+      alias: model.alias.trim(),
+      id: model.id.trim() === '' ? model.alias.trim() : model.id.trim(),
+      ...(model.category === undefined ? {} : { category: model.category }),
+    }))
     .filter(model => model.alias !== '')
   return {
     id: channel.id,

package/src/client/locales.ts

@@ -9,7 +9,8 @@ export const zh = {
   'mode.tts': '文本转语音',
   'mode.music': '音乐生成',
   'mode.sfx': '音效生成',
-  'prompt.placeholder': '输入要朗读的文本，或描述想生成的音乐 / 音效…',
+  'mode.voiceDesign': '音色设计',
+  'prompt.placeholder': '输入文本、音乐/音效描述，或音色设计描述…',
   'prompt.required': '请输入文本或提示词',
   'model.label': '模型 / 音色',
   'voice.label': '音色',
@@ -58,7 +59,7 @@ export const zh = {
   'channel.apiKey': 'API 密钥',
   'channel.apiKeyHint': '留空则保持当前密钥；输入新值可更换。',
   'channel.models': '模型 / 音色（每行一个：别名=上游ID）',
-  'channel.modelsHint': '例如 tts-1=tts-1 或 Rachel=21m00Tcm4TlvDq8ikWAM',
+  'channel.modelsHint': '每行格式：别名=上游ID，可加 @分类（tts/music/sfx/voice_design/voice_clone）；例如 tts-1=tts-1 @tts 或 Rachel=21m00Tcm4TlvDq8ikWAM @tts',
   'channel.default': '设为默认',
   'channel.cancel': '取消',
   'channel.save': '保存渠道',
@@ -75,7 +76,8 @@ export const en: Record<AudioGenKey, string> = {
   'mode.tts': 'Text to speech',
   'mode.music': 'Music',
   'mode.sfx': 'Sound effects',
-  'prompt.placeholder': 'Text to speak, or a description of the music / sound effect…',
+  'mode.voiceDesign': 'Voice design',
+  'prompt.placeholder': 'Text to speak, or a music/SFX/voice-design description…',
   'prompt.required': 'Prompt or text is required',
   'model.label': 'Model / voice',
   'voice.label': 'Voice',
@@ -124,7 +126,7 @@ export const en: Record<AudioGenKey, string> = {
   'channel.apiKey': 'API key',
   'channel.apiKeyHint': 'Leave blank to keep the current key.',
   'channel.models': 'Models / voices (one per line: alias=upstreamId)',
-  'channel.modelsHint': 'e.g. tts-1=tts-1 or Rachel=21m00Tcm4TlvDq8ikWAM',
+  'channel.modelsHint': 'One per line: alias=upstreamId, optional @category (tts/music/sfx/voice_design/voice_clone); e.g. tts-1=tts-1 @tts or Rachel=21m00Tcm4TlvDq8ikWAM @tts',
   'channel.default': 'Set default',
   'channel.cancel': 'Cancel',
   'channel.save': 'Save channel',

package/src/client/settings-scope.ts

@@ -14,7 +14,7 @@ import {
   type SettingsScopeSnapshot,
   type SnapshotStore,
 } from '@deepseek-ai/dsh-client-runtime/client'
-import { SETTINGS_API, type ChannelConfig } from '../protocol.ts'
+import { SETTINGS_API, type AudioModelCategory, type ChannelConfig } from '../protocol.ts'
 
 /** The fields this plugin's settings card edits. */
 export interface AudiogenConfig {
@@ -269,7 +269,10 @@ export function bindAudiogenScope(fetchFn: typeof fetch = fetch): AudiogenScope
  * Falls back to the legacy flat allow-list while no channels exist (upgrade
  * path). Pure projection — no host calls.
  */
-export function audioModelOptions(config: AudiogenConfig | undefined): { models: string[]; defaultChannelId?: string } {
+export function audioModelOptions(config: AudiogenConfig | undefined): {
+  models: Array<{ alias: string; category?: AudioModelCategory }>
+  defaultChannelId?: string
+} {
   const channels = config?.channels ?? []
   if (channels.length === 0) {
     return { models: [] }
@@ -278,11 +281,14 @@ export function audioModelOptions(config: AudiogenConfig | undefined): { models:
     ? config.defaultChannelId
     : channels[0]!.id
   const ordered = [defaultId, ...channels.filter(channel => channel.id !== defaultId).map(channel => channel.id)]
-  const models: string[] = []
+  const models: Array<{ alias: string; category?: AudioModelCategory }> = []
+  const seen = new Set<string>()
   for (const id of ordered) {
     const channel = channels.find(candidate => candidate.id === id)!
     for (const model of channel.models) {
-      if (model.alias !== '' && !models.includes(model.alias)) models.push(model.alias)
+      if (model.alias === '' || seen.has(model.alias)) continue
+      seen.add(model.alias)
+      models.push({ alias: model.alias, ...(model.category === undefined ? {} : { category: model.category }) })
     }
   }
   return models.length > 0 ? { models, defaultChannelId: defaultId } : { models: [], defaultChannelId: defaultId }

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 = '0.1.0'
+export const PLUGIN_VERSION = '0.3.0'
 
 /** Same-origin route family (loopback-only, mirroring dsh-imagegen). */
 export const SETTINGS_API = {
@@ -22,6 +22,11 @@ export const GENERATE_API = '/api/dsh-audiogen/generate' as const
 /** Host-mediated built-in provider catalog (channels the user can instantiate). */
 export const PRESETS_API = '/api/dsh-audiogen/presets' as const
 
+/** Host-mediated model/voice discovery endpoint. */
+export const MODEL_API = {
+  discover: '/api/dsh-audiogen/models/discover',
+} as const
+
 /** Loopback-only audio file reader for panel/tool-result previews. */
 export const AUDIO_API = {
   file: '/api/dsh-audiogen/audio',
@@ -40,7 +45,15 @@ export const HISTORY_API = {
 export const HISTORY_MAX = 50
 
 /** Audio generation modes. */
-export type AudioMode = 'tts' | 'music' | 'sfx'
+export type AudioMode = 'tts' | 'music' | 'sfx' | 'voice_design'
+
+/** The capability category of an audio model/voice. */
+export type AudioModelCategory =
+  | 'tts'
+  | 'music'
+  | 'sfx'
+  | 'voice_design'
+  | 'voice_clone'
 
 /** One model mapping in a channel's catalog: display alias → upstream id. */
 export interface ModelMapping {
@@ -48,6 +61,14 @@ export interface ModelMapping {
   alias: string
   /** Upstream model or voice id sent to the provider. */
   id: string
+  /** Optional capability category, used by the UI to group model lists. */
+  category?: AudioModelCategory
+}
+
+/** A model/voice discovered from a vendor endpoint (not yet persisted). */
+export interface DiscoveredAudioModel extends ModelMapping {
+  /** Optional human-readable description from the vendor. */
+  description?: string
 }
 
 /**
@@ -85,6 +106,8 @@ export interface GenerateAudioRequest {
   prompt: string
   /** Optional voice alias for TTS. */
   voice?: string
+  /** Optional preview text for voice-design APIs. */
+  previewText?: string
   /** Optional speaking rate / speed multiplier. */
   speed?: number
   /** Requested duration in seconds (music/sfx). */
@@ -113,6 +136,8 @@ export interface GeneratedAudio {
   url: string
   /** Stable audio id / file name. */
   id: string
+  /** Optional voice id returned by a voice-design API. */
+  voiceId?: string
 }
 
 /** Successful generate outcome. */
@@ -132,6 +157,8 @@ export interface HistoryAudioRef {
   mime: string
   /** Duration in seconds when known. */
   duration?: number
+  /** Optional generated voice id. */
+  voiceId?: string
 }
 
 /** A saved generation as the browser consumes it. */
@@ -142,6 +169,7 @@ export interface HistoryEntry {
   model: string
   prompt: string
   voice?: string
+  voiceId?: string
   speed?: number
   duration?: number
   format?: string
@@ -158,6 +186,7 @@ export interface HistoryEntryInput {
   model: string
   prompt: string
   voice?: string
+  voiceId?: string
   speed?: number
   duration?: number
   format?: string

package/src/routes.ts

@@ -11,10 +11,11 @@ import { randomUUID } from 'node:crypto'
 import type { WebRoute } from '@deepseek-ai/dsh-host-webserver'
 import { SettingsConflictError, settingsNamespace, type SettingsDescriptor } from '@deepseek-ai/dsh-settings'
 import { generateAudio, AudioGenError, type AudioChannel } from './audio-engine.ts'
+import { discoverAudioModels } from './audio-models.ts'
 import { AUDIO_PRESETS } from './audio-presets.ts'
 import { appendHistory, clearHistory, listHistory, readAudioFile, removeHistory, saveAudioFile } from './audio-store.ts'
 import {
-  AUDIO_API, AUDIOGEN_SETTINGS_NAMESPACE, GENERATE_API, HISTORY_API, PRESETS_API, SETTINGS_API,
+  AUDIO_API, AUDIOGEN_SETTINGS_NAMESPACE, GENERATE_API, HISTORY_API, MODEL_API, PRESETS_API, SETTINGS_API,
   type GenerateAudioRequest, type GeneratedAudio, type HistoryEntryInput,
 } from './protocol.ts'
 
@@ -89,7 +90,7 @@ function messageOf(error: unknown): string {
 }
 
 function parseGenerateRequest(body: Record<string, unknown>): GenerateAudioRequest | undefined {
-  const mode = body.mode === 'music' ? 'music' : body.mode === 'sfx' ? 'sfx' : 'tts'
+  const mode = body.mode === 'music' ? 'music' : body.mode === 'sfx' ? 'sfx' : body.mode === 'voice_design' ? 'voice_design' : 'tts'
   const prompt = typeof body.prompt === 'string' ? body.prompt.trim() : ''
   if (prompt === '') return undefined
   return {
@@ -97,6 +98,7 @@ function parseGenerateRequest(body: Record<string, unknown>): GenerateAudioReque
     model: typeof body.model === 'string' ? body.model.trim() : '',
     prompt,
     ...(typeof body.voice === 'string' && body.voice.trim() !== '' ? { voice: body.voice.trim() } : {}),
+    ...(typeof body.previewText === 'string' && body.previewText.trim() !== '' ? { previewText: body.previewText.trim() } : {}),
     ...(typeof body.speed === 'number' ? { speed: body.speed } : {}),
     ...(typeof body.duration === 'number' ? { duration: body.duration } : {}),
     ...(typeof body.format === 'string' && body.format.trim() !== '' ? { format: body.format.trim() } : {}),
@@ -140,6 +142,10 @@ function resolveChannelRequest(
   const target = explicit ?? defaults
   const asked = request.model.trim()
   if (asked === '') {
+    if (request.mode === 'voice_design') {
+      if (target === undefined) return { ok: false, code: 'no-channels', message: '尚未配置任何渠道' }
+      return { ok: true, request: { ...request, channelId: target.id, channel: target.name } }
+    }
     const alias = target?.models[0]?.alias ?? ''
     if (alias === '') {
       return { ok: false, code: 'no-models', message: `渠道「${target?.name ?? ''}」尚未配置模型/音色，请先在设置中添加` }
@@ -193,6 +199,32 @@ export function makeRoutes(deps: AudiogenRoutesDeps): WebRoute[] {
         writeJson(res, 200, { ok: true, presets: AUDIO_PRESETS })
       },
     },
+    // ---------------------------------------------- model/voice discovery
+    {
+      kind: 'exact',
+      path: MODEL_API.discover,
+      handler: async (req, res) => {
+        if (!guard(req, res, 'POST')) return
+        const body = await readJsonBody(req)
+        const view = deps.resolveChannels()
+        const stored = view.channels.find(candidate => candidate.id === (typeof body?.channelId === 'string' ? body.channelId : undefined))
+          ?? view.channels.find(candidate => candidate.id === view.defaultChannelId)
+          ?? view.channels[0]
+        const channel: AudioChannel = {
+          id: stored?.id ?? 'preview',
+          preset: stored?.preset ?? '',
+          name: stored?.name ?? '',
+          apiUrl: typeof body?.apiUrl === 'string' && body.apiUrl.trim() !== '' ? body.apiUrl.trim() : (stored?.apiUrl ?? ''),
+          apiKey: typeof body?.apiKey === 'string' && body.apiKey.trim() !== '' ? body.apiKey.trim() : (stored?.apiKey ?? ''),
+          models: stored?.models ?? [],
+        }
+        try {
+          writeJson(res, 200, { ok: true, ...await discoverAudioModels(channel) })
+        } catch (error) {
+          writeJson(res, 200, { ok: false, code: 'model-discovery-failed', message: messageOf(error) })
+        }
+      },
+    },
     // -------------------------------------------------- settings describe
     {
       kind: 'exact',
@@ -273,6 +305,7 @@ export function makeRoutes(deps: AudiogenRoutesDeps): WebRoute[] {
               mime: saved.mime,
               bytes: saved.bytes,
               url: `${AUDIO_API.file}/${encodeURIComponent(saved.file)}`,
+              ...(output.voiceId === undefined ? {} : { voiceId: output.voiceId }),
             })
           }
           let history