feat: add upstreamVoiceId to voice packs and related services

- Introduced upstreamVoiceId field in voice pack schema and database.
- Updated voice pack service to handle upstreamVoiceId in CRUD operations.
- Modified API routes and tests to accommodate upstreamVoiceId.
- Enhanced UI components to include upstreamVoiceId in forms and displays.
- Adjusted speech processing logic to utilize upstreamVoiceId where applicable.
- Updated related tests to ensure proper functionality with new field.

apps/server/drizzle/0016_tired_dagger.sql

@@ -0,0 +1,3 @@
+ALTER TABLE "voice_packs" ADD COLUMN "upstream_voice_id" text;
+UPDATE "voice_packs" SET "upstream_voice_id" = "voice_id";
+ALTER TABLE "voice_packs" ALTER COLUMN "upstream_voice_id" SET NOT NULL;

apps/server/drizzle/meta/_journal.json

@@ -113,6 +113,13 @@
       "when": 1780498188308,
       "tag": "0015_concerned_piledriver",
       "breakpoints": true
+    },
+    {
+      "idx": 16,
+      "version": "7",
+      "when": 1782847276369,
+      "tag": "0016_tired_dagger",
+      "breakpoints": true
     }
   ]
-}
+}

apps/server/src/routes/admin/voice-packs/route.test.ts

@@ -25,6 +25,7 @@ function createService() {
     provider: 'volcengine',
     model: 'seed-tts-2.0',
     voiceId: 'voice-neuro',
+    upstreamVoiceId: 'voice-neuro-upstream',
     ttsModelId: 'volcengine/neuro-pool',
     params: {},
     costMultiplier: 1.5,
@@ -119,6 +120,7 @@ describe('admin voice packs — CRUD', () => {
       provider: 'volcengine',
       model: 'seed-tts-2.0',
       voiceId: 'voice-neuro',
+      upstreamVoiceId: 'voice-neuro-upstream',
       ttsModelId: 'volcengine/neuro-pool',
       params: { pitch: '+20%' },
       costMultiplier: 1.5,
@@ -149,6 +151,7 @@ describe('admin voice packs — CRUD', () => {
       provider: 'volcengine',
       model: 'seed-tts-2.0',
       voiceId: 'voice-neuro',
+      upstreamVoiceId: 'voice-neuro-upstream',
       ttsModelId: 'volcengine/neuro-pool',
       params: {},
       costMultiplier: -1,

apps/server/src/routes/openai/v1/route.test.ts

@@ -710,7 +710,8 @@ describe('v1CompletionsRoutes', () => {
             description: null,
             provider: 'azure',
             model: 'microsoft/v1',
-            voiceId: 'en-US-AvaMultilingualNeural',
+            voiceId: 'friendly-azure',
+            upstreamVoiceId: 'en-US-AvaMultilingualNeural',
             ttsModelId: 'microsoft/v1',
             params: {},
             costMultiplier: 1.5,
@@ -728,12 +729,11 @@ describe('v1CompletionsRoutes', () => {
           body: JSON.stringify({
             model: 'auto',
             input: 'test',
-            voice: 'en-US-AvaMultilingualNeural',
+            voice: 'friendly-azure',
             speed: 1.2,
             extra_body: {
               voice_pack: {
                 pack_id: 'vp-azure',
-                cost_multiplier: 1.5,
                 pitch: 20,
                 volume: 5,
               },
@@ -784,7 +784,7 @@ describe('v1CompletionsRoutes', () => {
 
     /**
      * @example
-     * POST /api/v1/audio/speech { "input": "hello", "extra_body": { "voice_pack": { "cost_multiplier": 2 } } }
+     * POST /api/v1/audio/speech { "input": "hello", "extra_body": { "voice_pack": { "pack_id": "vp-premium" } } }
      */
     it('uses Voice Pack cost multiplier for affordability and billing units', async () => {
       globalThis.fetch = vi.fn(async () => new Response(new Uint8Array([1]), {
@@ -801,6 +801,7 @@ describe('v1CompletionsRoutes', () => {
           provider: 'azure',
           model: 'microsoft/v1',
           voiceId: 'alloy',
+          upstreamVoiceId: 'upstream-alloy',
           ttsModelId: 'tts-1',
           params: {},
           costMultiplier: 2,
@@ -832,7 +833,6 @@ describe('v1CompletionsRoutes', () => {
             extra_body: {
               voice_pack: {
                 pack_id: 'vp-premium',
-                cost_multiplier: 2,
               },
             },
           }),
@@ -868,6 +868,7 @@ describe('v1CompletionsRoutes', () => {
           provider: 'azure',
           model: 'microsoft/v1',
           voiceId: 'alloy',
+          upstreamVoiceId: 'upstream-alloy',
           ttsModelId: 'tts-1',
           params: {},
           costMultiplier: 2,
@@ -899,7 +900,6 @@ describe('v1CompletionsRoutes', () => {
             extra_body: {
               voice_pack: {
                 pack_id: 'vp-premium',
-                cost_multiplier: 2,
               },
               airi_analytics: {
                 source: 'manual_preview',

apps/server/src/routes/voice-packs/index.ts

@@ -5,6 +5,20 @@ import { Hono } from 'hono'
 
 import { authGuard } from '../../middlewares/auth'
 
+function publicVoicePack(pack: Awaited<ReturnType<VoicePackService['listEnabled']>>[number]) {
+  return {
+    id: pack.id,
+    name: pack.name,
+    description: pack.description,
+    voiceId: pack.voiceId,
+    params: pack.params,
+    costMultiplier: pack.costMultiplier,
+    enabled: pack.enabled,
+    createdAt: pack.createdAt,
+    updatedAt: pack.updatedAt,
+  }
+}
+
 /**
  * User-facing Voice Pack routes.
  *
@@ -17,6 +31,6 @@ export function createVoicePackRoutes(service: VoicePackService) {
     .use('*', authGuard)
     .get('/', async (c) => {
       const packs = await service.listEnabled()
-      return c.json(packs)
+      return c.json(packs.map(publicVoicePack))
     })
 }

apps/server/src/routes/voice-packs/route.test.ts

@@ -23,7 +23,21 @@ function createTestApp(service: VoicePackService, user: { id: string } | null) {
 
 function createService() {
   return {
-    listEnabled: vi.fn(async () => [{ id: 'vp-1', name: 'Enabled', enabled: true }]),
+    listEnabled: vi.fn(async () => [{
+      id: 'vp-1',
+      name: 'Enabled',
+      description: 'Public description',
+      provider: 'azure',
+      model: 'microsoft/v1',
+      voiceId: 'friendly-voice',
+      upstreamVoiceId: 'en-US-AvaMultilingualNeural',
+      ttsModelId: 'microsoft/v1',
+      params: { pitch: '+10%' },
+      costMultiplier: 2,
+      enabled: true,
+      createdAt: new Date('2026-01-01T00:00:00.000Z'),
+      updatedAt: new Date('2026-01-02T00:00:00.000Z'),
+    }]),
     list: vi.fn(),
     create: vi.fn(),
     update: vi.fn(),
@@ -50,7 +64,17 @@ describe('voice packs routes', () => {
     const res = await app.request('/api/v1/voice-packs')
 
     expect(res.status).toBe(200)
-    expect(await res.json()).toEqual([{ id: 'vp-1', name: 'Enabled', enabled: true }])
+    expect(await res.json()).toEqual([{
+      id: 'vp-1',
+      name: 'Enabled',
+      description: 'Public description',
+      voiceId: 'friendly-voice',
+      params: { pitch: '+10%' },
+      costMultiplier: 2,
+      enabled: true,
+      createdAt: '2026-01-01T00:00:00.000Z',
+      updatedAt: '2026-01-02T00:00:00.000Z',
+    }])
     expect(service.listEnabled).toHaveBeenCalled()
   })
 })

apps/server/src/schemas/voice-packs.ts

@@ -16,6 +16,7 @@ export const voicePacks = pgTable(
     provider: text('provider').notNull(),
     model: text('model').notNull(),
     voiceId: text('voice_id').notNull(),
+    upstreamVoiceId: text('upstream_voice_id').notNull(),
     ttsModelId: text('tts_model_id').notNull(),
     params: jsonb('params').notNull().$type<VoicePackParams>().default({}),
     costMultiplier: real('cost_multiplier').notNull().default(1),

apps/server/src/services/domain/openai-speech/index.ts

@@ -90,20 +90,23 @@ export function createOpenAiSpeechService(deps: OpenAiSpeechServiceDeps) {
 
   async function handleSpeechRequest(input: OpenAiSpeechRequest): Promise<Response> {
     const requestId = nanoid()
-    let requestModel = typeof input.body.model === 'string' ? input.body.model : 'auto'
+    const requestedModel = typeof input.body.model === 'string' ? input.body.model : 'auto'
+    let requestModel = requestedModel
+    const requestVoice = typeof input.body.voice === 'string' ? input.body.voice : undefined
     const inputText = typeof input.body.input === 'string' ? input.body.input : ''
     const analytics = ttsAnalyticsContext(input.body)
 
-    if (requestModel === 'auto')
-      requestModel = await deps.configKV.getOrThrow('DEFAULT_TTS_MODEL')
-
     const voicePackRequest = await voicePackRequestOptions(input.body, {
-      model: requestModel,
-      voice: typeof input.body.voice === 'string' ? input.body.voice : undefined,
+      requestedModel,
+      voice: requestVoice,
       voicePackService: deps.voicePackService,
     })
+    requestModel = voicePackRequest.model ?? requestModel
+    if (requestModel === 'auto')
+      requestModel = await deps.configKV.getOrThrow('DEFAULT_TTS_MODEL')
+    const routedVoice = voicePackRequest.voice ?? requestVoice
     const voiceMetadata = ttsVoiceMetadata({
-      voice: typeof input.body.voice === 'string' ? input.body.voice : undefined,
+      voice: requestVoice,
       voicePackId: voicePackRequest.voicePackId,
       voiceType: analytics.voiceType,
     })
@@ -114,7 +117,7 @@ export function createOpenAiSpeechService(deps: OpenAiSpeechServiceDeps) {
       userId: input.userId,
       model: requestModel,
       inputChars: inputText.length,
-      voice: typeof input.body.voice === 'string' ? input.body.voice : undefined,
+      voice: requestVoice,
     }).log('tts speech request')
 
     void deps.productEventService.track({
@@ -171,7 +174,7 @@ export function createOpenAiSpeechService(deps: OpenAiSpeechServiceDeps) {
 
     const ttsInput = {
       text: inputText,
-      voice: typeof input.body.voice === 'string' ? input.body.voice : undefined,
+      voice: routedVoice,
       speed: typeof input.body.speed === 'number' ? input.body.speed : undefined,
       responseFormat: typeof input.body.response_format === 'string' ? input.body.response_format : undefined,
       extraOptions: voicePackRequest.extraOptions,
@@ -405,19 +408,22 @@ function ttsVoiceMetadata(input: {
 async function voicePackRequestOptions(
   body: Record<string, unknown>,
   context: {
-    model: string
+    requestedModel: string
     voice?: string
     voicePackService: VoicePackService
   },
-): Promise<{ extraOptions: Record<string, unknown> | undefined, costMultiplier: number, voicePackId?: string }> {
+): Promise<{
+  extraOptions: Record<string, unknown> | undefined
+  costMultiplier: number
+  voicePackId?: string
+  model?: string
+  voice?: string
+}> {
   const extraBody = asRecord(body.extra_body)
   const voicePackOptions = asRecord(extraBody?.voice_pack)
   const pitch = readOptionalNumber(voicePackOptions, 'pitch')
   const volume = readOptionalNumber(voicePackOptions, 'volume')
-  const costMultiplier = await resolveVoicePackCostMultiplier(voicePackOptions, context)
-  const voicePackId = typeof voicePackOptions?.pack_id === 'string' && voicePackOptions.pack_id.trim()
-    ? voicePackOptions.pack_id
-    : undefined
+  const voicePack = await resolveVoicePackRequest(voicePackOptions, context)
   const extraOptions: Record<string, unknown> = {}
   if (pitch != null)
     extraOptions.pitch = pitch
@@ -426,40 +432,49 @@ async function voicePackRequestOptions(
 
   return {
     extraOptions: Object.keys(extraOptions).length > 0 ? extraOptions : undefined,
-    costMultiplier,
-    voicePackId,
+    costMultiplier: voicePack?.costMultiplier ?? 1,
+    voicePackId: voicePack?.id,
+    model: voicePack?.ttsModelId,
+    voice: voicePack?.upstreamVoiceId,
   }
 }
 
-async function resolveVoicePackCostMultiplier(
+async function resolveVoicePackRequest(
   voicePackOptions: Record<string, unknown> | undefined,
   context: {
-    model: string
+    requestedModel: string
     voice?: string
     voicePackService: VoicePackService
   },
-): Promise<number> {
+): Promise<Awaited<ReturnType<VoicePackService['findById']>> | null> {
   const packId = voicePackOptions?.pack_id
-  const value = voicePackOptions?.cost_multiplier
-  if (packId == null && value == null)
-    return 1
+  if (voicePackOptions?.cost_multiplier != null) {
+    throw createBadRequestError('voice_pack.cost_multiplier is server-managed', 'INVALID_VOICE_PACK', {
+      field: 'voice_pack.cost_multiplier',
+    })
+  }
+  if (packId == null)
+    return null
   if (typeof packId !== 'string' || !packId.trim())
     throw createBadRequestError('voice_pack.pack_id is required when Voice Pack billing metadata is provided', 'INVALID_VOICE_PACK')
 
   const pack = await context.voicePackService.findById(packId)
   if (!pack)
     throw createBadRequestError('Voice Pack not found', 'INVALID_VOICE_PACK', { packId })
-  if (pack.ttsModelId !== context.model || pack.voiceId !== context.voice) {
+  if (context.requestedModel !== 'auto' && pack.ttsModelId !== context.requestedModel) {
+    throw createBadRequestError('Voice Pack does not match requested model and voice', 'INVALID_VOICE_PACK', {
+      packId,
+      actualModel: context.requestedModel,
+    })
+  }
+  if (pack.voiceId !== context.voice) {
     throw createBadRequestError('Voice Pack does not match requested model and voice', 'INVALID_VOICE_PACK', {
       packId,
-      expectedModel: pack.ttsModelId,
-      actualModel: context.model,
-      expectedVoice: pack.voiceId,
       actualVoice: context.voice,
     })
   }
 
-  return pack.costMultiplier
+  return pack
 }
 
 function routerFailure(error: unknown): { status: number, reason: string, message: string } {

apps/server/src/services/domain/voice-packs/index.test.ts

@@ -27,6 +27,7 @@ describe('voicePackService', () => {
       provider: 'volcengine',
       model: 'seed-tts-2.0',
       voiceId: 'voice-neuro',
+      upstreamVoiceId: 'voice-neuro-upstream',
       ttsModelId: 'volcengine/neuro-pool',
       params: { pitch: '+20%', volume: '+5%' },
       costMultiplier: 1.5,
@@ -37,6 +38,7 @@ describe('voicePackService', () => {
     expect(pack.provider).toBe('volcengine')
     expect(pack.model).toBe('seed-tts-2.0')
     expect(pack.voiceId).toBe('voice-neuro')
+    expect(pack.upstreamVoiceId).toBe('voice-neuro-upstream')
     expect(pack.ttsModelId).toBe('volcengine/neuro-pool')
     expect(pack.params).toEqual({ pitch: '+20%', volume: '+5%' })
     expect(pack.costMultiplier).toBe(1.5)
@@ -50,6 +52,7 @@ describe('voicePackService', () => {
       provider: 'volcengine',
       model: 'seed-tts-2.0',
       voiceId: 'voice-a',
+      upstreamVoiceId: 'voice-a-upstream',
       ttsModelId: 'volcengine/pool',
       params: {},
       costMultiplier: 1,
@@ -60,6 +63,7 @@ describe('voicePackService', () => {
       provider: 'volcengine',
       model: 'seed-tts-2.0',
       voiceId: 'voice-a',
+      upstreamVoiceId: 'voice-a-upstream',
       ttsModelId: 'volcengine/pool',
       params: { pitch: '+20%' },
       costMultiplier: 1,
@@ -78,6 +82,7 @@ describe('voicePackService', () => {
       provider: 'azure',
       model: 'v1',
       voiceId: 'en-US-AvaMultilingualNeural',
+      upstreamVoiceId: 'en-US-AvaMultilingualNeural',
       ttsModelId: 'microsoft/v1',
       params: {},
       costMultiplier: 1,
@@ -103,6 +108,7 @@ describe('voicePackService', () => {
       provider: 'dashscope-cosyvoice',
       model: 'cosyvoice-v2',
       voiceId: 'longxiaochun_v2',
+      upstreamVoiceId: 'longxiaochun_v2',
       ttsModelId: 'alibaba/cosyvoice-v2',
       params: {},
       costMultiplier: 1,

apps/server/src/services/domain/voice-packs/index.ts

@@ -24,6 +24,7 @@ export const CreateVoicePackInputSchema = object({
   provider: pipe(string(), nonEmpty('provider is required'), maxLength(100)),
   model: pipe(string(), nonEmpty('model is required'), maxLength(200)),
   voiceId: pipe(string(), nonEmpty('voiceId is required'), maxLength(200)),
+  upstreamVoiceId: pipe(string(), nonEmpty('upstreamVoiceId is required'), maxLength(200)),
   ttsModelId: pipe(string(), nonEmpty('ttsModelId is required'), maxLength(200)),
   params: optional(VoicePackParamsSchema, {}),
   costMultiplier: VoicePackCostMultiplierSchema,
@@ -36,6 +37,7 @@ export const UpdateVoicePackInputSchema = object({
   provider: optional(pipe(string(), nonEmpty('provider must not be empty'), maxLength(100))),
   model: optional(pipe(string(), nonEmpty('model must not be empty'), maxLength(200))),
   voiceId: optional(pipe(string(), nonEmpty('voiceId must not be empty'), maxLength(200))),
+  upstreamVoiceId: optional(pipe(string(), nonEmpty('upstreamVoiceId must not be empty'), maxLength(200))),
   ttsModelId: optional(pipe(string(), nonEmpty('ttsModelId must not be empty'), maxLength(200))),
   params: optional(VoicePackParamsSchema),
   costMultiplier: optional(VoicePackCostMultiplierSchema),
@@ -74,6 +76,7 @@ export function createVoicePackService(db: Database) {
         provider: input.provider,
         model: input.model,
         voiceId: input.voiceId,
+        upstreamVoiceId: input.upstreamVoiceId,
         ttsModelId: input.ttsModelId,
         params: input.params,
         costMultiplier: input.costMultiplier,