dsh-audiogen 0.3.0 → 0.3.3

package/src/audio-models.ts

@@ -1,9 +1,17 @@
 /**
  * Host-side model/voice discovery.
  *
- * MiniMax exposes a voice-management API that returns all available system and
- * user-generated voice ids; we combine those with the known MiniMax music
- * models so the settings card can offer a full categorized catalog.
+ * Each vendor answers differently:
+ *  - MiniMax exposes a voice-management API (/v1/get_voice) that returns all
+ *    system and user-generated voice ids; the known MiniMax music models are
+ *    appended so the settings card can offer a full categorized catalog.
+ *  - ElevenLabs exposes /v1/models and /v1/voices; only models that can
+ *    actually speak (text_to_speech capability) and account voices are kept.
+ *  - Stability AI has no listing endpoint; the built-in stable-audio catalog
+ *    is returned.
+ *  - Generic OpenAI-compatible endpoints answer /models; the reply is filtered
+ *    to audio-related model ids only (tts / music / sfx), never the whole
+ *    model list of a gateway.
  */
 
 import type { AudioChannel } from './audio-engine.ts'
@@ -14,20 +22,38 @@ function isMiniMax(channel: AudioChannel): boolean {
   return channel.preset === 'minimax' || /minimax/i.test(channel.apiUrl)
 }
 
+function isElevenLabs(channel: AudioChannel): boolean {
+  return channel.preset === 'elevenlabs' || /elevenlabs/i.test(channel.apiUrl)
+}
+
+function isStability(channel: AudioChannel): boolean {
+  return channel.preset === 'stability' || /stability\.ai/i.test(channel.apiUrl)
+}
+
 function baseUrl(url: string): string {
   return url.trim().replace(/\/+$/, '')
 }
 
+/** Whether an upstream model id is audio-related at all. */
 function categoryFor(id: string): AudioModelCategory | undefined {
   const value = id.toLowerCase()
-  if (/(tts|speech|voice|t2a)/i.test(value)) return 'tts'
-  if (/(music|song|cover|lyrics)/i.test(value)) return 'music'
+  if (/(tts|speech|voice|t2a|talk|narration)/i.test(value)) return 'tts'
+  if (/(music|song|cover|lyrics|audio|melody|beat)/i.test(value)) return 'music'
   if (/(sfx|sound.?effect|effect|foley)/i.test(value)) return 'sfx'
   return undefined
 }
 
+async function fetchJson(url: string, init: RequestInit): Promise<unknown> {
+  const response = await fetch(url, init)
+  if (!response.ok) {
+    const text = await response.text().catch(() => '')
+    throw new Error(`HTTP ${response.status}${text === '' ? '' : `: ${text.slice(0, 300)}`}`)
+  }
+  return response.json()
+}
+
 async function postJson(url: string, apiKey: string, body: unknown): Promise<unknown> {
-  const response = await fetch(url, {
+  return fetchJson(url, {
     method: 'POST',
     headers: {
       authorization: `Bearer ${apiKey.trim()}`,
@@ -35,11 +61,6 @@ async function postJson(url: string, apiKey: string, body: unknown): Promise<unk
     },
     body: JSON.stringify(body),
   })
-  if (!response.ok) {
-    const text = await response.text().catch(() => '')
-    throw new Error(`HTTP ${response.status}${text === '' ? '' : `: ${text.slice(0, 300)}`}`)
-  }
-  return response.json()
 }
 
 /** Discover available models/voices for a channel. */
@@ -47,9 +68,18 @@ export async function discoverAudioModels(channel: AudioChannel): Promise<{ mode
   if (channel.apiUrl.trim() === '') throw new Error('API URL is not configured')
   if (channel.apiKey.trim() === '') throw new Error('API key is not configured')
 
-  if (isMiniMax(channel)) {
-    const base = baseUrl(channel.apiUrl).replace(/\/v1$/i, '')
-    const url = `${base}/v1/get_voice`
+  if (isMiniMax(channel)) return discoverMiniMax(channel)
+  if (isElevenLabs(channel)) return discoverElevenLabs(channel)
+  if (isStability(channel)) return discoverStability(channel)
+  return discoverOpenAICompatible(channel)
+}
+
+// ---------------------------------------------------------------- MiniMax
+
+async function discoverMiniMax(channel: AudioChannel): Promise<{ models: DiscoveredAudioModel[]; source: string }> {
+  const base = baseUrl(channel.apiUrl).replace(/\/v1$/i, '')
+  const url = `${base}/v1/get_voice`
+  try {
     const payload = await postJson(url, channel.apiKey, { voice_type: 'all' }) as {
       system_voice?: Array<{ voice_id?: string; voice_name?: string; description?: string[] }>
       voice_cloning?: Array<{ voice_id?: string; description?: string[] }>
@@ -94,27 +124,105 @@ export async function discoverAudioModels(channel: AudioChannel): Promise<{ mode
     const music = (audioPresetById('minimax')?.models ?? []).filter(model => model.category === 'music')
     for (const model of music) models.push({ ...model, category: 'music' as const })
     const deduped = dedupe(models)
-    return { models: deduped, source: 'MiniMax get_voice + built-in music catalog' }
+    return { models: deduped, source: 'MiniMax get_voice + music 目录' }
+  } catch (error) {
+    // Gateways may not route /v1/get_voice (or lack voice-management access).
+    // Fall back to the built-in catalog so generation still works.
+    const fallback = (audioPresetById('minimax')?.models ?? []).map(model => ({ ...model }))
+    const message = error instanceof Error ? error.message : String(error)
+    return {
+      models: dedupe(fallback),
+      source: `内置 MiniMax 目录（音色发现失败：${message.slice(0, 160)}）`,
+    }
   }
+}
 
-  // Best-effort OpenAI-compatible /models discovery.
+// ------------------------------------------------------------- ElevenLabs
+
+async function discoverElevenLabs(channel: AudioChannel): Promise<{ models: DiscoveredAudioModel[]; source: string }> {
+  const base = baseUrl(channel.apiUrl)
+  const headers = { 'xi-api-key': channel.apiKey.trim() }
+  const failures: string[] = []
+  const models: DiscoveredAudioModel[] = []
+
+  // 1. TTS-capable models from /v1/models (audio-related only).
+  try {
+    const payload = await fetchJson(`${base}/models`, { headers }) as Array<{
+      model_id?: string
+      name?: string
+      description?: string
+      capabilities?: { text_to_speech?: boolean; voice_change?: boolean; speech_to_text?: boolean }
+    }>
+    for (const item of Array.isArray(payload) ? payload : []) {
+      const id = item.model_id?.trim() ?? ''
+      if (id === '') continue
+      // Only what actually produces speech audio.
+      if (item.capabilities?.text_to_speech !== true && item.capabilities?.voice_change !== true) continue
+      models.push({
+        alias: item.name?.trim() || id,
+        id,
+        category: 'tts',
+        ...(item.description !== undefined && item.description.trim() !== '' ? { description: item.description.trim() } : {}),
+      })
+    }
+  } catch (error) {
+    failures.push(`模型列表：${error instanceof Error ? error.message : String(error)}`)
+  }
+
+  // 2. The account's voices from /v1/voices, grouped as tts entries.
+  try {
+    const payload = await fetchJson(`${base}/voices`, { headers }) as {
+      voices?: Array<{ voice_id?: string; name?: string; description?: string }>
+    }
+    for (const voice of Array.isArray(payload?.voices) ? payload.voices : []) {
+      const id = voice.voice_id?.trim() ?? ''
+      if (id === '') continue
+      models.push({
+        alias: voice.name?.trim() || id,
+        id,
+        category: 'tts',
+        ...(voice.description !== undefined && voice.description.trim() !== '' ? { description: voice.description.trim() } : {}),
+      })
+    }
+  } catch (error) {
+    failures.push(`音色列表：${error instanceof Error ? error.message : String(error)}`)
+  }
+
+  if (models.length === 0) {
+    // Neither endpoint answered — fall back to the built-in catalog.
+    const fallback = (audioPresetById('elevenlabs')?.models ?? []).map(model => ({ ...model }))
+    const detail = failures.length === 0 ? '' : `（发现失败：${failures.join('；').slice(0, 160)}）`
+    return { models: dedupe(fallback), source: `内置 ElevenLabs 目录${detail}` }
+  }
+  return { models: dedupe(models), source: 'ElevenLabs /models + /voices' }
+}
+
+// -------------------------------------------------------------- Stability
+
+async function discoverStability(channel: AudioChannel): Promise<{ models: DiscoveredAudioModel[]; source: string }> {
+  // Stability has no public audio model listing; serve the built-in catalog.
+  const fallback = (audioPresetById('stability-audio')?.models ?? []).map(model => ({ ...model }))
+  return { models: dedupe(fallback), source: 'Stability stable-audio 内置目录' }
+}
+
+// ------------------------------------------------------ OpenAI-compatible
+
+async function discoverOpenAICompatible(channel: AudioChannel): Promise<{ models: DiscoveredAudioModel[]; source: string }> {
   const base = baseUrl(channel.apiUrl)
   const url = `${base}/models`
-  const response = await fetch(url, {
+  const payload = await fetchJson(url, {
     headers: { authorization: `Bearer ${channel.apiKey.trim()}` },
-  })
-  if (!response.ok) {
-    throw new Error(`model list request failed (HTTP ${response.status}); please add models manually`)
-  }
-  const payload = await response.json() as { data?: Array<{ id?: string }> }
+  }) as { data?: Array<{ id?: string }> }
   const models: DiscoveredAudioModel[] = []
   for (const item of payload.data ?? []) {
     const id = item.id?.trim() ?? ''
     if (id === '') continue
-    const category = categoryFor(id) ?? 'tts'
+    // Audio-related models only — never the whole gateway model list.
+    const category = categoryFor(id)
+    if (category === undefined) continue
     models.push({ alias: id, id, category })
   }
-  return { models: dedupe(models), source: 'OpenAI-compatible /models' }
+  return { models: dedupe(models), source: 'OpenAI-compatible /models（仅音频相关）' }
 }
 
 function dedupe(models: DiscoveredAudioModel[]): DiscoveredAudioModel[] {

package/src/audio-presets.ts

@@ -1,6 +1,10 @@
 /**
  * Built-in audio provider catalog (presets).
  * Framework-free pure data; served to the settings card through a host route.
+ *
+ * Only the officially supported audio vendors are offered here — MiniMax,
+ * ElevenLabs and Stability AI. Any other endpoint (including OpenAI-compatible
+ * TTS gateways) is added through the "自定义渠道" flow instead.
  */
 
 import type { ModelMapping } from './protocol.ts'
@@ -15,39 +19,19 @@ export interface AudioPresetProvider {
   apiUrl: string
   /** One-line description shown in the picker. */
   hint: string
+  /** Official vendor website, shown as a link in the channel editor. */
+  site?: string
   /** Known model/voice catalog prefilled into the channel. */
   models: ModelMapping[]
 }
 
 export const AUDIO_PRESETS: AudioPresetProvider[] = [
-  {
-    id: 'openai-tts',
-    name: 'OpenAI · TTS',
-    apiUrl: 'https://api.openai.com/v1',
-    hint: 'OpenAI 官方语音合成接口（/audio/speech）',
-    models: [
-      { alias: 'tts-1', id: 'tts-1', category: 'tts' },
-      { alias: 'tts-1-hd', id: 'tts-1-hd', category: 'tts' },
-      { alias: 'gpt-4o-mini-tts', id: 'gpt-4o-mini-tts', category: 'tts' },
-    ],
-  },
-  {
-    id: 'elevenlabs',
-    name: 'ElevenLabs',
-    apiUrl: 'https://api.elevenlabs.io/v1',
-    hint: 'ElevenLabs TTS；模型列表请填写你的 Voice ID（如 Rachel / Adam 等别名）',
-    models: [
-      { alias: 'Rachel', id: '21m00Tcm4TlvDq8ikWAM', category: 'tts' },
-      { alias: 'Adam', id: 'pNInz6obpgDQGcFmaJgB', category: 'tts' },
-      { alias: 'Antoni', id: 'ErXwobaYiN019PkySvjV', category: 'tts' },
-      { alias: 'Bella', id: 'EXAVITQu4vr4xnSDxMaL', category: 'tts' },
-    ],
-  },
   {
     id: 'minimax',
     name: 'MiniMax',
     apiUrl: 'https://api.minimaxi.com',
-    hint: 'MiniMax 音色设计 / TTS / 音乐生成；可使用“获取可用模型”拉取账号音色',
+    site: 'https://www.minimaxi.com',
+    hint: 'MiniMax 官方音频：音色设计 / TTS / 音乐生成；建议点击「获取可用模型」拉取账号音色与模型',
     models: [
       // TTS models
       { alias: 'speech-2.8-hd', id: 'speech-2.8-hd', category: 'tts' },
@@ -64,23 +48,33 @@ export const AUDIO_PRESETS: AudioPresetProvider[] = [
       { alias: 'music-cover', id: 'music-cover', category: 'music' },
     ],
   },
+  {
+    id: 'elevenlabs',
+    name: 'ElevenLabs',
+    apiUrl: 'https://api.elevenlabs.io/v1',
+    site: 'https://elevenlabsai.cn',
+    hint: 'ElevenLabs 语音合成（TTS）；建议点击「获取可用模型」拉取音色与模型',
+    models: [
+      { alias: 'Rachel', id: '21m00Tcm4TlvDq8ikWAM', category: 'tts' },
+      { alias: 'Adam', id: 'pNInz6obpgDQGcFmaJgB', category: 'tts' },
+      { alias: 'Antoni', id: 'ErXwobaYiN019PkySvjV', category: 'tts' },
+      { alias: 'Bella', id: 'EXAVITQu4vr4xnSDxMaL', category: 'tts' },
+      { alias: 'eleven_multilingual_v2', id: 'eleven_multilingual_v2', category: 'tts' },
+      { alias: 'eleven_turbo_v2_5', id: 'eleven_turbo_v2_5', category: 'tts' },
+      { alias: 'eleven_flash_v2_5', id: 'eleven_flash_v2_5', category: 'tts' },
+    ],
+  },
   {
     id: 'stability-audio',
-    name: 'Stability AI · 音频',
+    name: 'Stability AI（stable-audio）',
     apiUrl: 'https://api.stability.ai/v2beta/audio',
-    hint: 'Stability AI 音乐/音效生成（stable-audio 系列）',
+    site: 'https://stability.ai/stable-audio',
+    hint: 'Stability AI 音乐 / 音效生成（stable-audio 系列）',
     models: [
       { alias: 'stable-audio-2.0', id: 'stable-audio-2.0', category: 'music' },
       { alias: 'stable-audio-1.0', id: 'stable-audio-1.0', category: 'music' },
     ],
   },
-  {
-    id: 'custom',
-    name: '自定义渠道',
-    apiUrl: '',
-    hint: '任意兼容接口；支持 OpenAI 兼容 TTS，或返回音频字节 / JSON 的通用 POST',
-    models: [],
-  },
 ]
 
 /** Look up one built-in provider by id. */

package/src/client/AudioGenPanel.tsx

@@ -60,12 +60,28 @@ export function AudioGenPanel(props: { api: AudiogenApi; scope: AudiogenScope })
   const [voice, setVoice] = useState('')
   const [speed, setSpeed] = useState('')
   const [duration, setDuration] = useState('')
+  const [lyrics, setLyrics] = useState('')
+  const [instrumental, setInstrumental] = useState(false)
   const [format, setFormat] = useState('mp3')
+  // MiniMax TTS 高级参数（其他厂商忽略）
+  const [emotion, setEmotion] = useState('')
+  const [vol, setVol] = useState('')
+  const [pitch, setPitch] = useState('')
+  const [toneText, setToneText] = useState('')
+  const [sampleRate, setSampleRate] = useState('')
+  const [bitrate, setBitrate] = useState('')
+  const [audioChannel, setAudioChannel] = useState('')
+  const [subtitle, setSubtitle] = useState(false)
   const [loading, setLoading] = useState(false)
   const [error, setError] = useState<string | null>(null)
   const [outputs, setOutputs] = useState<GeneratedAudio[]>([])
   const { entries, reload, clear } = useHistory()
 
+  const isMiniMaxChannel = useMemo(() => {
+    const target = channels.find(candidate => candidate.id === modelOptions.defaultChannelId) ?? channels[0]
+    return target !== undefined && (target.preset === 'minimax' || /minimax/i.test(target.apiUrl))
+  }, [channels, modelOptions.defaultChannelId])
+
   const visibleModels = useMemo(() => {
     if (mode === 'voice_design') return []
     return modelOptions.models
@@ -95,7 +111,17 @@ export function AudioGenPanel(props: { api: AudiogenApi; scope: AudiogenScope })
         ...(voice.trim() !== '' ? { voice: voice.trim() } : {}),
         ...(speed.trim() !== '' ? { speed: Number(speed) } : {}),
         ...(duration.trim() !== '' ? { duration: Number(duration) } : {}),
+        ...(lyrics.trim() !== '' ? { lyrics: lyrics.trim() } : {}),
+        ...(instrumental ? { isInstrumental: true } : {}),
         ...(format.trim() !== '' ? { format: format.trim() } : {}),
+        ...(emotion.trim() !== '' ? { emotion: emotion.trim() } : {}),
+        ...(vol.trim() !== '' ? { vol: Number(vol) } : {}),
+        ...(pitch.trim() !== '' ? { pitch: Number(pitch) } : {}),
+        ...(toneText.trim() !== '' ? { pronunciationTone: toneText.split('\n').map(item => item.trim()).filter(item => item !== '') } : {}),
+        ...(sampleRate.trim() !== '' ? { sampleRate: Number(sampleRate) } : {}),
+        ...(bitrate.trim() !== '' ? { bitrate: Number(bitrate) } : {}),
+        ...(audioChannel.trim() !== '' ? { audioChannel: Number(audioChannel) } : {}),
+        ...(subtitle ? { subtitleEnable: true } : {}),
       })
       if (!response.ok) {
         setError(response.message ?? '生成失败')
@@ -166,7 +192,7 @@ export function AudioGenPanel(props: { api: AudiogenApi; scope: AudiogenScope })
           {mode === 'tts' ? (
             <label className={css.label}>
               <span>{tt('voice.label')}</span>
-              <input className={css.input} value={voice} onChange={event => setVoice(event.target.value)} placeholder="alloy / 自定义音色" />
+              <input className={css.input} value={voice} onChange={event => setVoice(event.target.value)} placeholder={isMiniMaxChannel ? 'male-qn-qingse / female-shaonv' : 'alloy / 自定义音色'} />
             </label>
           ) : null}
 
@@ -177,6 +203,52 @@ export function AudioGenPanel(props: { api: AudiogenApi; scope: AudiogenScope })
             </label>
           ) : null}
 
+          {mode === 'tts' && isMiniMaxChannel ? (
+            <details className={css.advanced}>
+              <summary>MiniMax 高级参数</summary>
+              <label className={css.label}>
+                <span>情绪 emotion</span>
+                <input className={css.input} value={emotion} onChange={event => setEmotion(event.target.value)} placeholder="happy / sad / angry / nervous…" />
+              </label>
+              <div className={css.row}>
+                <label className={css.label}>
+                  <span>音量 vol (0-10)</span>
+                  <input className={css.input} type="number" min="0" max="10" step="0.5" value={vol} onChange={event => setVol(event.target.value)} placeholder="1" />
+                </label>
+                <label className={css.label}>
+                  <span>音调 pitch (-12~12)</span>
+                  <input className={css.input} type="number" min="-12" max="12" value={pitch} onChange={event => setPitch(event.target.value)} placeholder="0" />
+                </label>
+              </div>
+              <div className={css.row}>
+                <label className={css.label}>
+                  <span>采样率</span>
+                  <input className={css.input} type="number" min="16000" max="48000" step="8000" value={sampleRate} onChange={event => setSampleRate(event.target.value)} placeholder="32000" />
+                </label>
+                <label className={css.label}>
+                  <span>码率 bps</span>
+                  <input className={css.input} type="number" min="64000" max="320000" step="8000" value={bitrate} onChange={event => setBitrate(event.target.value)} placeholder="128000" />
+                </label>
+                <label className={css.label}>
+                  <span>声道</span>
+                  <select className={css.select} value={audioChannel} onChange={event => setAudioChannel(event.target.value)}>
+                    <option value="">默认(1)</option>
+                    <option value="1">1</option>
+                    <option value="2">2</option>
+                  </select>
+                </label>
+              </div>
+              <label className={css.label}>
+                <span>发音词典（每行一条："文字/读音"）</span>
+                <textarea className={css.textarea} value={toneText} onChange={event => setToneText(event.target.value)} placeholder={'处理/(chu3)(li3)\n危险/dangerous'} />
+              </label>
+              <label className={css.checkbox}>
+                <input type="checkbox" checked={subtitle} onChange={event => setSubtitle(event.target.checked)} />
+                <span>生成字幕 subtitle_enable</span>
+              </label>
+            </details>
+          ) : null}
+
           {mode === 'music' || mode === 'sfx' ? (
             <label className={css.label}>
               <span>{tt('duration.label')}</span>
@@ -184,14 +256,31 @@ export function AudioGenPanel(props: { api: AudiogenApi; scope: AudiogenScope })
             </label>
           ) : null}
 
+          {mode === 'music' ? (
+            <>
+              <label className={css.label}>
+                <span>歌词（纯音乐模式可留空；多段用空行分隔）</span>
+                <textarea className={css.textarea} value={lyrics} onChange={event => setLyrics(event.target.value)} placeholder={'第一段歌词…\n\n第二段歌词…'} />
+              </label>
+              <label className={css.checkbox}>
+                <input type="checkbox" checked={instrumental} onChange={event => setInstrumental(event.target.checked)} />
+                <span>纯音乐（无歌词/人声）is_instrumental</span>
+              </label>
+            </>
+          ) : null}
+
           {needModel ? (
             <label className={css.label}>
               <span>{tt('format.label')}</span>
               <select className={css.select} value={format} onChange={event => setFormat(event.target.value)}>
                 <option value="mp3">mp3</option>
                 <option value="wav">wav</option>
-                <option value="flac">flac</option>
-                <option value="ogg">ogg</option>
+                {mode === 'tts' ? (
+                  <>
+                    <option value="flac">flac</option>
+                    <option value="ogg">ogg</option>
+                  </>
+                ) : null}
                 <option value="pcm">pcm</option>
               </select>
             </label>