@h-ai/ai 0.1.0-alpha.49 → 0.1.0-alpha.50

package/README.md

@@ -213,6 +213,8 @@ if (setup.success) {
 
 先在 `ai.init()` 中注册语音模型并映射默认识别/合成模型。凭据默认回退到平台环境变量；只有 LLM 与语音模型确认使用同一凭据时，才显式启用 `inheritLlmApiKey`：
 
+平台（`provider`）表示调用协议而非部署位置：`openai` / `mimo` / `qwen` / `doubao` 为云服务；`whisper`（ASR）/ `indextts`（TTS）为可自托管协议，无 canonical 端点，必须显式配置 `baseUrl`，同一协议可切换多个 Endpoint（云 / 内网 / 本地）而不改业务代码。云平台缺少凭据返回 `CONFIGURATION_ERROR`；自托管平台可无凭据（配置 `apiKey` 后按 `Authorization: Bearer` 发送）。
+
 ```ts
 await ai.init({
   llm: {
@@ -223,18 +225,42 @@ await ai.init({
     models: [
       { id: 'asr', provider: 'qwen', model: 'qwen3-asr-flash-realtime', operations: ['transcribe'] },
       { id: 'tts', provider: 'qwen', model: 'qwen3-tts-flash-realtime', operations: ['synthesize'] },
+      // 自托管：whisper / indextts 必须显式配置 baseUrl
+      { id: 'whisper', provider: 'whisper', model: 'faster-whisper-large-v3', operations: ['transcribe'], baseUrl: 'http://127.0.0.1:8101/v1' },
+      { id: 'indextts', provider: 'indextts', model: 'indextts-2.5', operations: ['synthesize'], baseUrl: 'http://127.0.0.1:8102/v1' },
     ],
     transcribeModel: 'asr',
     synthesizeModel: 'tts',
   },
 })
 
-// 完整识别（可选热词提示提升专有名词识别率）
-const result = await ai.audio.transcribe({ audio: { data: wavBytes, format: 'wav' }, language: 'zh', contextHints: ['专有名词'] })
+// 完整识别（可选热词、时间戳粒度与 VAD；strictCapabilities 要求模型真实支持所请求能力）
+const result = await ai.audio.transcribe({
+  audio: { data: wavBytes, format: 'wav' },
+  language: 'zh',
+  contextHints: ['专有名词'],
+  timestampGranularities: ['segment', 'word'],
+  vad: true,
+  model: 'whisper',
+})
 if (result.success) {
   const text = result.data.text
+  const language = result.data.language // 实际检测语言
+  const words = result.data.segments?.[0]?.words // 词级时间轴（毫秒整数）
 }
 
+// 说话人 / 风格参考合成（IndexTTS）：speakerReference 表达“谁在说”，styleReference 表达“怎么说”
+const cloned = await ai.audio.synthesize({
+  text: '我们下午三点出发。',
+  language: 'zh',
+  model: 'indextts',
+  speakerReference: { audio: speakerWav, language: 'ja' },
+  styleReference: { audio: styleWav },
+  styleStrength: 0.8,
+  targetDurationMs: 3280, // 目标时长（与 speed 互斥），metadata.durationMatched 反馈是否达标
+  durationToleranceMs: 120,
+})
+
 // 实时识别（持续音频输入 → 领域事件流：speech_started / transcript / speech_stopped）
 for await (const event of ai.audio.transcribeStream({
   audio: { chunks: microphoneChunks, format: 'pcm16', sampleRate: 16000 },
@@ -290,6 +316,8 @@ return playable.data
 
 浏览器 / 移动端通过 `@h-ai/serv` 暴露的统一语音 WebSocket 入口访问，`@h-ai/ai/client` 提供与 Node 端一致的 `audio.*` API（传输细节内部隐藏）。浏览器客户端严格区分正常结束、取消（`AUDIO_CANCELLED`）与异常断连（`AUDIO_CONNECTION_FAILED`）：取消或在 `end` 前断连会抛出对应领域错误码，`synthesize` 不会把未完成的部分音频当作成功结果返回。
 
+> 自托管模型服务（faster-whisper / IndexTTS / Qwen3-4B 的 CPU/GPU Docker 镜像、权重下载与离线打包）见 [`models/`](./models/README.md)。镜像与权重下载优先使用 ModelScope（中国网络友好），自动回退 HuggingFace 镜像。
+
 ### 文生图（Image）
 
 在 `ai.init()` 注册模型后，调用方只接收标准化图片字节，不感知厂商 Base64、内联数据或临时 URL：

package/dist/{ai-audio-ws-protocol-BAOyKbBU.d.ts → ai-audio-ws-protocol-COIRfUFF.d.ts}

@@ -1,6 +1,6 @@
 import * as _h_ai_core from '@h-ai/core';
 import { HaiResult } from '@h-ai/core';
-import { aQ as ChatMessage, ap as InteractionScope, U as MemoryType, bB as RagOptions, bF as ReasoningOptions, cc as ToolRegistryOperations, bn as MemoryEntry, bU as SessionInfo, bg as LLMOperations, bt as MemoryOperations, bA as RagOperations, bE as ReasoningOperations, c as AIConfig, d as AIConfigInput, aF as AIStoreProvider, ce as ToolsOperations, bY as StreamOperations, bL as RetrievalOperations, ba as KnowledgeOperations, az as A2AOperations, q as AudioOperations, k as AudioContent, $ as PlayableAudio, l as AudioFormat } from './ai-reasoning-types-DwxulYKu.js';
+import { a_ as ChatMessage, az as InteractionScope, Y as MemoryType, bL as RagOptions, bP as ReasoningOptions, cm as ToolRegistryOperations, bx as MemoryEntry, c2 as SessionInfo, bq as LLMOperations, bD as MemoryOperations, bK as RagOperations, bO as ReasoningOperations, c as AIConfig, d as AIConfigInput, aP as AIStoreProvider, co as ToolsOperations, c6 as StreamOperations, bV as RetrievalOperations, bk as KnowledgeOperations, aJ as A2AOperations, s as AudioOperations, k as AudioContent, a3 as PlayableAudio, l as AudioFormat } from './ai-reasoning-types-CuITylkA.js';
 import * as zod from 'zod';
 import { z } from 'zod';
 import { Buffer } from 'node:buffer';

package/dist/{ai-reasoning-types-DwxulYKu.d.ts → ai-reasoning-types-CuITylkA.d.ts}

@@ -2759,21 +2759,43 @@ type A2AConfig = z.infer<typeof A2AConfigSchema>;
 /**
  * 语音平台枚举
  *
- * 决定 `ai.audio` 底层调用哪个厂商（对使用方透明，公共请求/响应形状保持一致）：
+ * 决定 `ai.audio` 底层调用哪个厂商 / 协议（对使用方透明，公共请求/响应形状保持一致）：
  *
  * - `openai` — OpenAI Audio API（transcriptions / speech）
  * - `mimo` — 小米 MiMo（Chat Completions 风格 ASR / TTS）
  * - `qwen` — 阿里云百炼 Qwen Realtime（DashScope WebSocket ASR / TTS）
  * - `doubao` — 火山引擎豆包语音（二进制 WebSocket ASR / TTS）
+ * - `whisper` — hai-framework Whisper Service 协议（HTTP 文件识别，可自托管）
+ * - `indextts` — hai-framework IndexTTS Service 协议（HTTP 合成，可自托管）
+ *
+ * 平台仅表示「用哪种协议调用」，不表示服务部署在哪里；`whisper` / `indextts` 需显式配置 `baseUrl`。
  */
 declare const AudioProviderSchema: z.ZodEnum<{
     openai: "openai";
     mimo: "mimo";
     qwen: "qwen";
     doubao: "doubao";
+    whisper: "whisper";
+    indextts: "indextts";
 }>;
 /** 语音平台类型 */
 type AudioProviderName = z.infer<typeof AudioProviderSchema>;
+/** 单个语音操作类型 */
+declare const AudioOperationSchema: z.ZodEnum<{
+    transcribe: "transcribe";
+    synthesize: "synthesize";
+}>;
+/** 语音操作类型 */
+type AudioOperation = z.infer<typeof AudioOperationSchema>;
+/**
+ * 语音模型允许的操作列表
+ *
+ * 至少一项，且不允许重复。
+ */
+declare const AudioOperationsSchema: z.ZodArray<z.ZodEnum<{
+    transcribe: "transcribe";
+    synthesize: "synthesize";
+}>>;
 /**
  * 语音模型条目 Schema
  *
@@ -2791,9 +2813,14 @@ declare const AudioModelEntrySchema: z.ZodObject<{
         mimo: "mimo";
         qwen: "qwen";
         doubao: "doubao";
+        whisper: "whisper";
+        indextts: "indextts";
     }>;
     model: z.ZodString;
-    operations: z.ZodUnion<readonly [z.ZodTuple<[z.ZodLiteral<"transcribe">], null>, z.ZodTuple<[z.ZodLiteral<"synthesize">], null>, z.ZodTuple<[z.ZodLiteral<"transcribe">, z.ZodLiteral<"synthesize">], null>]>;
+    operations: z.ZodArray<z.ZodEnum<{
+        transcribe: "transcribe";
+        synthesize: "synthesize";
+    }>>;
     apiKey: z.ZodOptional<z.ZodPipe<z.ZodNullable<z.ZodString>, z.ZodTransform<string | undefined, string | null>>>;
     baseUrl: z.ZodOptional<z.ZodString>;
     appKey: z.ZodOptional<z.ZodPipe<z.ZodNullable<z.ZodString>, z.ZodTransform<string | undefined, string | null>>>;
@@ -2831,9 +2858,14 @@ declare const AudioConfigSchema: z.ZodObject<{
             mimo: "mimo";
             qwen: "qwen";
             doubao: "doubao";
+            whisper: "whisper";
+            indextts: "indextts";
         }>;
         model: z.ZodString;
-        operations: z.ZodUnion<readonly [z.ZodTuple<[z.ZodLiteral<"transcribe">], null>, z.ZodTuple<[z.ZodLiteral<"synthesize">], null>, z.ZodTuple<[z.ZodLiteral<"transcribe">, z.ZodLiteral<"synthesize">], null>]>;
+        operations: z.ZodArray<z.ZodEnum<{
+            transcribe: "transcribe";
+            synthesize: "synthesize";
+        }>>;
         apiKey: z.ZodOptional<z.ZodPipe<z.ZodNullable<z.ZodString>, z.ZodTransform<string | undefined, string | null>>>;
         baseUrl: z.ZodOptional<z.ZodString>;
         appKey: z.ZodOptional<z.ZodPipe<z.ZodNullable<z.ZodString>, z.ZodTransform<string | undefined, string | null>>>;
@@ -2890,6 +2922,17 @@ interface ResolvedAudioModel {
  * @returns 成功返回已解析模型；无匹配模型返回 `AUDIO_MODEL_NOT_FOUND`；缺少凭据返回 `CONFIGURATION_ERROR`
  */
 declare function resolveAudioModel(audioConfig: AudioConfig, operation: 'transcribe' | 'synthesize', explicit?: string, llmApiKey?: string): HaiResult<ResolvedAudioModel>;
+/**
+ * 校验已解析模型是否具备该平台协议所需的凭据
+ *
+ * 凭据是否必需由平台协议决定：自托管的 `whisper` / `indextts` 可无凭据；豆包允许 API Key
+ * 或 App/Access Key 组合；其余云平台必须提供 API Key。仅在真正发起识别 / 合成前调用，
+ * 能力查询（`getCapabilities`）不需要凭据。
+ *
+ * @param model - 已解析模型
+ * @returns 缺少必需凭据时返回 `CONFIGURATION_ERROR`，否则返回 `null`
+ */
+declare function ensureAudioCredential(model: ResolvedAudioModel): HaiError | null;
 /**
  * 文生图平台枚举
  *
@@ -3175,9 +3218,14 @@ declare const AIConfigSchema: z.ZodObject<{
                 mimo: "mimo";
                 qwen: "qwen";
                 doubao: "doubao";
+                whisper: "whisper";
+                indextts: "indextts";
             }>;
             model: z.ZodString;
-            operations: z.ZodUnion<readonly [z.ZodTuple<[z.ZodLiteral<"transcribe">], null>, z.ZodTuple<[z.ZodLiteral<"synthesize">], null>, z.ZodTuple<[z.ZodLiteral<"transcribe">, z.ZodLiteral<"synthesize">], null>]>;
+            operations: z.ZodArray<z.ZodEnum<{
+                transcribe: "transcribe";
+                synthesize: "synthesize";
+            }>>;
             apiKey: z.ZodOptional<z.ZodPipe<z.ZodNullable<z.ZodString>, z.ZodTransform<string | undefined, string | null>>>;
             baseUrl: z.ZodOptional<z.ZodString>;
             appKey: z.ZodOptional<z.ZodPipe<z.ZodNullable<z.ZodString>, z.ZodTransform<string | undefined, string | null>>>;
@@ -3274,10 +3322,46 @@ interface AudioInputStream {
     /** 声道数（默认单声道） */
     channels?: 1 | 2;
 }
-/** 完整语音识别请求 */
-interface TranscriptionRequest {
-    /** 待识别的完整音频 */
-    audio: AudioContent;
+/**
+ * ASR 时间戳粒度
+ *
+ * - `segment` — 语义段 / 句级时间轴
+ * - `word` — 词级时间轴
+ */
+type TranscriptionTimestampGranularity = 'segment' | 'word';
+/** 单个 ASR 词级识别结果 */
+interface TranscriptionWord {
+    /** 识别出的文字、单词或 Token */
+    text: string;
+    /** 起始时间（毫秒） */
+    startMs: number;
+    /** 结束时间（毫秒） */
+    endMs: number;
+    /** 置信度（仅 Provider 能提供明确概率语义时返回） */
+    confidence?: number;
+}
+/** ASR 分段结果 */
+interface TranscriptionSegment {
+    /** Provider 返回或 Framework 生成的 Segment ID */
+    id?: string;
+    /** 当前 Segment 完整文本 */
+    text: string;
+    /** Segment 起始位置（毫秒） */
+    startMs: number;
+    /** Segment 结束位置（毫秒） */
+    endMs: number;
+    /** 当前 Segment 下的词级时间轴 */
+    words?: TranscriptionWord[];
+    /** 为后续 Speaker Diarization 预留的说话人 ID */
+    speakerId?: string;
+}
+/**
+ * ASR 公共请求选项
+ *
+ * 完整识别与流式识别共用；所有高级能力字段均为可选，不支持的平台会忽略或（在
+ * `strictCapabilities` 为真时）提前失败。
+ */
+interface TranscriptionOptions {
     /** 语言提示（如 `zh` / `en`；不传时由模型自动检测） */
     language?: string;
     /**
@@ -3286,32 +3370,41 @@ interface TranscriptionRequest {
      * Provider 按能力映射为热词表 / phrase list / vocabulary / 提示词；不支持的平台会忽略。
      */
     contextHints?: string[];
+    /** 请求返回的时间戳粒度（`segment` / `word`） */
+    timestampGranularities?: TranscriptionTimestampGranularity[];
+    /** 是否启用模型 / 服务端 VAD */
+    vad?: boolean;
+    /** 是否严格要求高级能力全部被支持（true 时不支持则提前失败，false 为 best effort） */
+    strictCapabilities?: boolean;
     /** 模型 ID（不传时使用配置中的默认识别模型） */
     model?: string;
     /** 取消信号 */
     signal?: AbortSignal;
 }
+/** 完整语音识别请求 */
+interface TranscriptionRequest extends TranscriptionOptions {
+    /** 待识别的完整音频 */
+    audio: AudioContent;
+}
 /** 流式语音识别请求（支持完整音频或持续音频输入） */
-interface TranscriptionStreamRequest {
+interface TranscriptionStreamRequest extends TranscriptionOptions {
     /** 完整音频，或持续到达的音频输入流 */
     audio: AudioContent | AudioInputStream;
-    /** 语言提示（如 `zh` / `en`；不传时由模型自动检测） */
-    language?: string;
-    /**
-     * 领域提示词 / 热词（如角色名、专有名词、当前主题关键词）
-     *
-     * Provider 按能力映射为热词表 / phrase list / vocabulary / 提示词；不支持的平台会忽略。
-     */
-    contextHints?: string[];
-    /** 模型 ID（不传时使用配置中的默认识别模型） */
-    model?: string;
-    /** 取消信号 */
-    signal?: AbortSignal;
 }
 /** 完整语音识别结果 */
 interface TranscriptionResult {
     /** 识别文本 */
     text: string;
+    /** 实际检测或使用的语言 */
+    language?: string;
+    /** 输入音频总时长（毫秒） */
+    durationMs?: number;
+    /**
+     * 结构化识别时间轴
+     *
+     * Word Timestamp 统一存储在 `segments[].words`，不在顶层重复。
+     */
+    segments?: TranscriptionSegment[];
 }
 /**
  * 流式语音识别领域事件
@@ -3330,30 +3423,79 @@ type TranscriptionEvent = {
     type: 'transcript';
     text: string;
     final: boolean;
+    startMs?: number;
+    endMs?: number;
+    words?: TranscriptionWord[];
 } | {
     type: 'speech_stopped';
 };
-/** 完整语音合成请求 */
-interface SynthesisRequest {
-    /** 待合成文本 */
-    text: string;
+/**
+ * TTS 通用参考音频
+ *
+ * 不绑定任何具体 TTS 模型。既可表达「谁在说话」（说话人参考），也可表达「怎么说」（风格参考），
+ * 由 `SynthesisOptions.speakerReference` / `styleReference` 决定用途。
+ */
+interface AudioReference {
+    /** 参考音频 */
+    audio: AudioContent;
+    /** 参考音频对应文本（仅需要参考 Transcript 的 Provider 使用） */
+    transcript?: string;
+    /** 参考音频语言 */
+    language?: string;
+}
+/**
+ * TTS 公共请求选项
+ *
+ * 完整合成与流式合成共用；所有高级能力字段均为可选，不支持的平台会忽略或（在
+ * `strictCapabilities` 为真时）提前失败。
+ */
+interface SynthesisOptions {
+    /** 目标语言 */
+    language?: string;
     /** 音色（厂商音色名，不传时使用模型默认音色） */
     voice?: string;
+    /** 说话人 / 音色参考（表达「谁在说话」） */
+    speakerReference?: AudioReference;
+    /** 风格、情绪、韵律参考（表达「怎么说」） */
+    styleReference?: AudioReference;
+    /** 风格参考影响强度，统一为 `[0, 1]` */
+    styleStrength?: number;
     /**
      * 自然语言风格指令（如语速、情绪、角色语气）
      *
      * Provider 按能力映射（如 MiMo 放入 user 消息、Qwen instructions）；不支持的平台会忽略。
      */
     instruction?: string;
+    /**
+     * 语速倍数
+     *
+     * `1.0` 正常、`>1` 更快、`<1` 更慢。
+     */
+    speed?: number;
     /** 输出音频格式（不传时使用模型默认格式） */
     format?: AudioFormat;
     /** 输出采样率（Hz） */
     sampleRate?: number;
     /** 模型 ID（不传时使用配置中的默认合成模型） */
     model?: string;
+    /** 是否严格要求高级能力全部被支持（true 时不支持则提前失败，false 为 best effort） */
+    strictCapabilities?: boolean;
     /** 取消信号 */
     signal?: AbortSignal;
 }
+/** 完整语音合成请求 */
+interface SynthesisRequest extends SynthesisOptions {
+    /** 待合成文本 */
+    text: string;
+    /**
+     * 最终音频目标时长（毫秒）
+     *
+     * 表达 Framework 业务目标，不对应某个模型私有参数；由支持的 Provider / Model Service 闭环逼近。
+     */
+    targetDurationMs?: number;
+    /** 目标时长允许误差（毫秒） */
+    durationToleranceMs?: number;
+}
 /** 带稳定 ID 的合成文本段 */
 interface SynthesisTextSegment {
     /** 调用方分配的稳定 ID，用于关联文本、音频与播放完成状态 */
@@ -3362,28 +3504,21 @@ interface SynthesisTextSegment {
     text: string;
 }
 /** 流式语音合成请求（支持单段或持续文本段输入） */
-interface SynthesisStreamRequest {
+interface SynthesisStreamRequest extends SynthesisOptions {
     /** 单个文本段，或持续到达的文本段流 */
     text: SynthesisTextSegment | AsyncIterable<SynthesisTextSegment>;
-    /** 音色（厂商音色名，不传时使用模型默认音色） */
-    voice?: string;
-    /**
-     * 自然语言风格指令（如语速、情绪、角色语气）
-     *
-     * Provider 按能力映射（如 MiMo 放入 user 消息、Qwen instructions）；不支持的平台会忽略。
-     */
-    instruction?: string;
-    /** 输出音频格式（不传时使用模型默认格式） */
-    format?: AudioFormat;
-    /** 输出采样率（Hz） */
-    sampleRate?: number;
-    /** 模型 ID（不传时使用配置中的默认合成模型） */
-    model?: string;
-    /** 取消信号 */
-    signal?: AbortSignal;
 }
 /** 完整语音合成结果 */
 interface SynthesisResult extends AudioContent {
+    /** 实际输出音频时长（毫秒） */
+    durationMs?: number;
+    /** 通用生成元数据 */
+    metadata?: {
+        /** 是否满足调用方指定的时长容差（无法判断时保持 undefined） */
+        durationMatched?: boolean;
+        /** Provider 最终使用的 Framework speed */
+        speed?: number;
+    };
 }
 /**
  * 流式语音合成领域事件
@@ -3420,16 +3555,55 @@ type SynthesisEvent = {
 interface AudioModelCapabilities {
     /** 语音识别能力；模型未声明识别操作时不返回 */
     transcribe?: {
+        /** 是否支持 ASR */
         supported: boolean;
+        /** 是否原生支持持续实时音频输入 */
         realtimeAudioInput: boolean;
+        /** 是否产生语音边界事件（speech_started / speech_stopped） */
         speechBoundaryEvents: boolean;
+        /** 是否原生输出增量 Transcript */
         streamingTranscriptOutput: boolean;
+        /** 是否接受 Language Hint */
+        languageHint?: boolean;
+        /** 是否支持自动语言检测 */
+        languageDetection?: boolean;
+        /** 是否支持 Segment Timestamp */
+        segmentTimestamps?: boolean;
+        /** 是否支持 Word Timestamp */
+        wordTimestamps?: boolean;
+        /** 是否支持 Context Hints */
+        contextHints?: boolean;
+        /** 是否支持 VAD */
+        vad?: boolean;
+        /** 是否支持 Speaker Diarization */
+        speakerDiarization?: boolean;
     };
     /** 语音合成能力；模型未声明合成操作时不返回 */
     synthesize?: {
+        /** 是否支持 TTS */
         supported: boolean;
+        /** 是否原生支持增量文本输入 */
         incrementalTextInput: boolean;
+        /** 是否原生支持流式音频输出 */
         streamingAudioOutput: boolean;
+        /** 是否支持指定目标语言 */
+        languageSelection?: boolean;
+        /** 是否支持预置 Voice */
+        presetVoice?: boolean;
+        /** 是否支持 Speaker Reference */
+        speakerReference?: boolean;
+        /** 是否必须提供 Speaker Reference */
+        speakerReferenceRequired?: boolean;
+        /** 是否支持 Style Reference */
+        styleReference?: boolean;
+        /** 是否支持通用自然语言 Instruction */
+        instruction?: boolean;
+        /** 是否支持 Framework Speed */
+        speedControl?: boolean;
+        /** 是否支持 Framework Target Duration */
+        targetDuration?: boolean;
+        /** 已知支持语言 */
+        supportedLanguages?: string[];
     };
 }
 /** 查询语音模型能力的参数 */
@@ -3756,4 +3930,4 @@ interface ReasoningOperations {
     runStream: (query: string, options?: ReasoningOptions) => AsyncIterable<ReasoningStreamEvent>;
 }
 
-export { type PlayableAudio as $, type A2AConfig as A, ImageModelEntrySchema as B, type CompressConfig as C, type ImageProviderName as D, type EmbeddingConfig as E, type FileConfig as F, ImageProviderSchema as G, KnowledgeConfigSchema as H, type ImageConfig as I, LLMConfigSchema as J, type KnowledgeConfig as K, type LLMConfig as L, type MCPConfig as M, MCPConfigSchema as N, type MCPServerCapabilities as O, MCPServerCapabilitiesSchema as P, type MCPServerConfig as Q, MCPServerConfigSchema as R, type MemoryConfig as S, MemoryConfigSchema as T, type MemoryType as U, MemoryTypeSchema as V, type ModelEntry as W, ModelEntrySchema as X, type ModelScenario as Y, ModelScenarioSchema as Z, OptionalSecretSchema as _, A2AConfigSchema as a, type KnowledgeAskOptions as a$, type ResolveRequiredModelEntryOptions as a0, type ResolvedAudioModel as a1, type ResolvedImageModel as a2, type ResolvedModelConfig as a3, type RetrievalConfig as a4, RetrievalConfigSchema as a5, type RetrievalSourceConfig as a6, RetrievalSourceSchema as a7, type SummaryConfig as a8, SummaryConfigSchema as a9, type A2ASecurityConfig as aA, type A2ATaskFilter as aB, type AILLMFunctionsDeps as aC, type AIRelStore as aD, type AIRelStoreOptions as aE, type AIStoreProvider as aF, type AIVectorBackend as aG, type AIVectorStore as aH, type AskOptions as aI, type AssistantMessage as aJ, type ChatCompletionChoice as aK, type ChatCompletionChunk as aL, type ChatCompletionDelta as aM, type ChatCompletionRequest as aN, type ChatCompletionResponse as aO, type ChatHistoryOptions as aP, type ChatMessage as aQ, type ChatRecord as aR, type Citation as aS, type DefineToolOptions as aT, type DeveloperMessage as aU, type EntityDocumentRelation as aV, type EntityDocumentResult as aW, type EntityListOptions as aX, type EntityQueryOptions as aY, type GenerateObjectRequest as aZ, type ImageContent as a_, type SynthesisEvent as aa, type SynthesisRequest as ab, type SynthesisResult as ac, type SynthesisStreamRequest as ad, type SynthesisTextSegment as ae, type TokenConfig as af, TokenConfigSchema as ag, type TranscriptionEvent as ah, type TranscriptionRequest as ai, type TranscriptionResult as aj, type TranscriptionStreamRequest as ak, resolveAudioModel as al, resolveImageModel as am, resolveModelApi as an, resolveModelEntry as ao, type InteractionScope as ap, type A2AAgentCardConfig as aq, type A2AApiKeySecurity as ar, type A2AAuthenticator as as, type A2ACallOptions as at, type A2ACallResult as au, type A2ACallerIdentity as av, type A2AContextInfo as aw, type A2AHandleResult as ax, type A2AMessageRecord as ay, type A2AOperations as az, A2ASkillConfigSchema as b, type SystemMessage as b$, type KnowledgeAskResult as b0, type KnowledgeDocumentInfo as b1, type KnowledgeDocumentListOptions as b2, type KnowledgeDocumentRemoveOptions as b3, type KnowledgeEntity as b4, type KnowledgeIngestBatchProgress as b5, type KnowledgeIngestBatchResult as b6, type KnowledgeIngestFileInput as b7, type KnowledgeIngestInput as b8, type KnowledgeIngestResult as b9, type RagOperations as bA, type RagOptions as bB, type RagResult as bC, type RagStreamEvent as bD, type ReasoningOperations as bE, type ReasoningOptions as bF, type ReasoningResult as bG, type ReasoningStep as bH, type ReasoningStepType as bI, type ReasoningStrategy as bJ, type ReasoningStreamEvent as bK, type RetrievalOperations as bL, type RetrievalRequest as bM, type RetrievalResult as bN, type RetrievalResultItem as bO, type RetrievalSource as bP, type SSEDecoder as bQ, type SSEEvent as bR, type ScopedMemoryBinding as bS, type ScopedMemoryOperations as bT, type SessionInfo as bU, type StoreFilter as bV, type StorePage as bW, type StoreScope as bX, type StreamOperations as bY, type StreamProcessor as bZ, type StreamResult as b_, type KnowledgeOperations as ba, type KnowledgeRetrieveItem as bb, type KnowledgeRetrieveOptions as bc, type KnowledgeRetrieveResult as bd, type KnowledgeSetupOptions as be, type KnowledgeStore as bf, type LLMOperations as bg, type LLMProvider as bh, type MemoryAccessScope as bi, type MemoryAdminOperations as bj, type MemoryClearAllOptions as bk, type MemoryClearOptions as bl, type MemoryCoreOperations as bm, type MemoryEntry as bn, type MemoryEntryInput as bo, type MemoryExtractOptions as bp, type MemoryInjectionOptions as bq, type MemoryListOptions as br, type MemoryListPageOptions as bs, type MemoryOperations as bt, type MemoryRecallOptions as bu, type MemoryUpdateInput as bv, type MessageContent as bw, type MessageRole as bx, type ObjectRef as by, type RagContextItem as bz, type AIConfig as c, type TempModelConfig as c0, type TextContent as c1, type TokenUsage as c2, type Tool as c3, type ToolAuthorizationRequest as c4, type ToolAuthorizer as c5, type ToolCall as c6, type ToolDefinition as c7, type ToolErrorType as c8, type ToolExecutionContext as c9, type ToolExecutionOptions as ca, type ToolMessage as cb, type ToolRegistryOperations as cc, type ToolRegistryOptions as cd, type ToolsOperations as ce, type UserMessage as cf, type WhereClause as cg, type WhereOperator as ch, type WhereValue as ci, type AIConfigInput as d, AIConfigSchema as e, type ApiType as f, ApiTypeSchema as g, type AudioCapabilitiesRequest as h, type AudioConfig as i, AudioConfigSchema as j, type AudioContent as k, type AudioFormat as l, type AudioInputStream as m, type AudioModelCapabilities as n, type AudioModelEntry as o, AudioModelEntrySchema as p, type AudioOperations as q, type AudioProviderName as r, AudioProviderSchema as s, CompressConfigSchema as t, EmbeddingConfigSchema as u, type EntityType as v, EntityTypeSchema as w, FileConfigSchema as x, ImageConfigSchema as y, type ImageModelEntry as z };
+export { ModelEntrySchema as $, type A2AConfig as A, EntityTypeSchema as B, type CompressConfig as C, FileConfigSchema as D, type EmbeddingConfig as E, type FileConfig as F, ImageConfigSchema as G, type ImageModelEntry as H, type ImageConfig as I, ImageModelEntrySchema as J, type ImageProviderName as K, ImageProviderSchema as L, type KnowledgeConfig as M, KnowledgeConfigSchema as N, type LLMConfig as O, LLMConfigSchema as P, type MCPConfig as Q, MCPConfigSchema as R, type MCPServerCapabilities as S, MCPServerCapabilitiesSchema as T, type MCPServerConfig as U, MCPServerConfigSchema as V, type MemoryConfig as W, MemoryConfigSchema as X, type MemoryType as Y, MemoryTypeSchema as Z, type ModelEntry as _, A2AConfigSchema as a, type ChatRecord as a$, type ModelScenario as a0, ModelScenarioSchema as a1, OptionalSecretSchema as a2, type PlayableAudio as a3, type ResolveRequiredModelEntryOptions as a4, type ResolvedAudioModel as a5, type ResolvedImageModel as a6, type ResolvedModelConfig as a7, type RetrievalConfig as a8, RetrievalConfigSchema as a9, type A2AAgentCardConfig as aA, type A2AApiKeySecurity as aB, type A2AAuthenticator as aC, type A2ACallOptions as aD, type A2ACallResult as aE, type A2ACallerIdentity as aF, type A2AContextInfo as aG, type A2AHandleResult as aH, type A2AMessageRecord as aI, type A2AOperations as aJ, type A2ASecurityConfig as aK, type A2ATaskFilter as aL, type AILLMFunctionsDeps as aM, type AIRelStore as aN, type AIRelStoreOptions as aO, type AIStoreProvider as aP, type AIVectorBackend as aQ, type AIVectorStore as aR, type AskOptions as aS, type AssistantMessage as aT, type ChatCompletionChoice as aU, type ChatCompletionChunk as aV, type ChatCompletionDelta as aW, type ChatCompletionRequest as aX, type ChatCompletionResponse as aY, type ChatHistoryOptions as aZ, type ChatMessage as a_, type RetrievalSourceConfig as aa, RetrievalSourceSchema as ab, type SummaryConfig as ac, SummaryConfigSchema as ad, type SynthesisEvent as ae, type SynthesisOptions as af, type SynthesisRequest as ag, type SynthesisResult as ah, type SynthesisStreamRequest as ai, type SynthesisTextSegment as aj, type TokenConfig as ak, TokenConfigSchema as al, type TranscriptionEvent as am, type TranscriptionOptions as an, type TranscriptionRequest as ao, type TranscriptionResult as ap, type TranscriptionSegment as aq, type TranscriptionStreamRequest as ar, type TranscriptionTimestampGranularity as as, type TranscriptionWord as at, ensureAudioCredential as au, resolveAudioModel as av, resolveImageModel as aw, resolveModelApi as ax, resolveModelEntry as ay, type InteractionScope as az, A2ASkillConfigSchema as b, type SSEEvent as b$, type Citation as b0, type DefineToolOptions as b1, type DeveloperMessage as b2, type EntityDocumentRelation as b3, type EntityDocumentResult as b4, type EntityListOptions as b5, type EntityQueryOptions as b6, type GenerateObjectRequest as b7, type ImageContent as b8, type KnowledgeAskOptions as b9, type MemoryInjectionOptions as bA, type MemoryListOptions as bB, type MemoryListPageOptions as bC, type MemoryOperations as bD, type MemoryRecallOptions as bE, type MemoryUpdateInput as bF, type MessageContent as bG, type MessageRole as bH, type ObjectRef as bI, type RagContextItem as bJ, type RagOperations as bK, type RagOptions as bL, type RagResult as bM, type RagStreamEvent as bN, type ReasoningOperations as bO, type ReasoningOptions as bP, type ReasoningResult as bQ, type ReasoningStep as bR, type ReasoningStepType as bS, type ReasoningStrategy as bT, type ReasoningStreamEvent as bU, type RetrievalOperations as bV, type RetrievalRequest as bW, type RetrievalResult as bX, type RetrievalResultItem as bY, type RetrievalSource as bZ, type SSEDecoder as b_, type KnowledgeAskResult as ba, type KnowledgeDocumentInfo as bb, type KnowledgeDocumentListOptions as bc, type KnowledgeDocumentRemoveOptions as bd, type KnowledgeEntity as be, type KnowledgeIngestBatchProgress as bf, type KnowledgeIngestBatchResult as bg, type KnowledgeIngestFileInput as bh, type KnowledgeIngestInput as bi, type KnowledgeIngestResult as bj, type KnowledgeOperations as bk, type KnowledgeRetrieveItem as bl, type KnowledgeRetrieveOptions as bm, type KnowledgeRetrieveResult as bn, type KnowledgeSetupOptions as bo, type KnowledgeStore as bp, type LLMOperations as bq, type LLMProvider as br, type MemoryAccessScope as bs, type MemoryAdminOperations as bt, type MemoryClearAllOptions as bu, type MemoryClearOptions as bv, type MemoryCoreOperations as bw, type MemoryEntry as bx, type MemoryEntryInput as by, type MemoryExtractOptions as bz, type AIConfig as c, type ScopedMemoryBinding as c0, type ScopedMemoryOperations as c1, type SessionInfo as c2, type StoreFilter as c3, type StorePage as c4, type StoreScope as c5, type StreamOperations as c6, type StreamProcessor as c7, type StreamResult as c8, type SystemMessage as c9, type TempModelConfig as ca, type TextContent as cb, type TokenUsage as cc, type Tool as cd, type ToolAuthorizationRequest as ce, type ToolAuthorizer as cf, type ToolCall as cg, type ToolDefinition as ch, type ToolErrorType as ci, type ToolExecutionContext as cj, type ToolExecutionOptions as ck, type ToolMessage as cl, type ToolRegistryOperations as cm, type ToolRegistryOptions as cn, type ToolsOperations as co, type UserMessage as cp, type WhereClause as cq, type WhereOperator as cr, type WhereValue as cs, type AIConfigInput as d, AIConfigSchema as e, type ApiType as f, ApiTypeSchema as g, type AudioCapabilitiesRequest as h, type AudioConfig as i, AudioConfigSchema as j, type AudioContent as k, type AudioFormat as l, type AudioInputStream as m, type AudioModelCapabilities as n, type AudioModelEntry as o, AudioModelEntrySchema as p, type AudioOperation as q, AudioOperationSchema as r, type AudioOperations as s, AudioOperationsSchema as t, type AudioProviderName as u, AudioProviderSchema as v, type AudioReference as w, CompressConfigSchema as x, EmbeddingConfigSchema as y, type EntityType as z };

package/dist/browser.d.ts

@@ -1,5 +1,5 @@
-export { A as A2AConfig, a as A2AConfigSchema, b as A2ASkillConfigSchema, c as AIConfig, d as AIConfigInput, e as AIConfigSchema, f as ApiType, g as ApiTypeSchema, h as AudioCapabilitiesRequest, i as AudioConfig, j as AudioConfigSchema, k as AudioContent, l as AudioFormat, m as AudioInputStream, n as AudioModelCapabilities, o as AudioModelEntry, p as AudioModelEntrySchema, q as AudioOperations, r as AudioProviderName, s as AudioProviderSchema, C as CompressConfig, t as CompressConfigSchema, E as EmbeddingConfig, u as EmbeddingConfigSchema, v as EntityType, w as EntityTypeSchema, F as FileConfig, x as FileConfigSchema, I as ImageConfig, y as ImageConfigSchema, z as ImageModelEntry, B as ImageModelEntrySchema, D as ImageProviderName, G as ImageProviderSchema, K as KnowledgeConfig, H as KnowledgeConfigSchema, L as LLMConfig, J as LLMConfigSchema, M as MCPConfig, N as MCPConfigSchema, O as MCPServerCapabilities, P as MCPServerCapabilitiesSchema, Q as MCPServerConfig, R as MCPServerConfigSchema, S as MemoryConfig, T as MemoryConfigSchema, U as MemoryType, V as MemoryTypeSchema, W as ModelEntry, X as ModelEntrySchema, Y as ModelScenario, Z as ModelScenarioSchema, _ as OptionalSecretSchema, $ as PlayableAudio, a0 as ResolveRequiredModelEntryOptions, a1 as ResolvedAudioModel, a2 as ResolvedImageModel, a3 as ResolvedModelConfig, a4 as RetrievalConfig, a5 as RetrievalConfigSchema, a6 as RetrievalSourceConfig, a7 as RetrievalSourceSchema, a8 as SummaryConfig, a9 as SummaryConfigSchema, aa as SynthesisEvent, ab as SynthesisRequest, ac as SynthesisResult, ad as SynthesisStreamRequest, ae as SynthesisTextSegment, af as TokenConfig, ag as TokenConfigSchema, ah as TranscriptionEvent, ai as TranscriptionRequest, aj as TranscriptionResult, ak as TranscriptionStreamRequest, al as resolveAudioModel, am as resolveImageModel, an as resolveModelApi, ao as resolveModelEntry } from './ai-reasoning-types-DwxulYKu.js';
-export { A as AIFunctions, a as AIInitOptions, b as AUDIO_WS_PATH, c as AudioFormatSchema, d as AudioWsClientMessage, e as AudioWsClientMessageSchema, f as AudioWsDoneMessage, g as AudioWsDoneMessageSchema, h as AudioWsEndMessage, i as AudioWsErrorMessage, j as AudioWsSegmentDoneMessage, k as AudioWsSegmentStartedMessage, l as AudioWsServerMessage, m as AudioWsSpeechMessage, n as AudioWsStartMessage, o as AudioWsStartMessageSchema, p as AudioWsTextMessage, q as AudioWsTextMessageSchema, r as AudioWsTranscriptMessage, C as CompressionStrategy, s as CompressionStrategySchema, G as GenerateImageRequest, t as GenerateImageResult, u as GeneratedImage, H as HaiAIError, I as ImageOperations, v as ImageSize, R as ReferenceImage, w as serializePlayableAudio } from './ai-audio-ws-protocol-BAOyKbBU.js';
+export { A as A2AConfig, a as A2AConfigSchema, b as A2ASkillConfigSchema, c as AIConfig, d as AIConfigInput, e as AIConfigSchema, f as ApiType, g as ApiTypeSchema, h as AudioCapabilitiesRequest, i as AudioConfig, j as AudioConfigSchema, k as AudioContent, l as AudioFormat, m as AudioInputStream, n as AudioModelCapabilities, o as AudioModelEntry, p as AudioModelEntrySchema, q as AudioOperation, r as AudioOperationSchema, s as AudioOperations, t as AudioOperationsSchema, u as AudioProviderName, v as AudioProviderSchema, w as AudioReference, C as CompressConfig, x as CompressConfigSchema, E as EmbeddingConfig, y as EmbeddingConfigSchema, z as EntityType, B as EntityTypeSchema, F as FileConfig, D as FileConfigSchema, I as ImageConfig, G as ImageConfigSchema, H as ImageModelEntry, J as ImageModelEntrySchema, K as ImageProviderName, L as ImageProviderSchema, M as KnowledgeConfig, N as KnowledgeConfigSchema, O as LLMConfig, P as LLMConfigSchema, Q as MCPConfig, R as MCPConfigSchema, S as MCPServerCapabilities, T as MCPServerCapabilitiesSchema, U as MCPServerConfig, V as MCPServerConfigSchema, W as MemoryConfig, X as MemoryConfigSchema, Y as MemoryType, Z as MemoryTypeSchema, _ as ModelEntry, $ as ModelEntrySchema, a0 as ModelScenario, a1 as ModelScenarioSchema, a2 as OptionalSecretSchema, a3 as PlayableAudio, a4 as ResolveRequiredModelEntryOptions, a5 as ResolvedAudioModel, a6 as ResolvedImageModel, a7 as ResolvedModelConfig, a8 as RetrievalConfig, a9 as RetrievalConfigSchema, aa as RetrievalSourceConfig, ab as RetrievalSourceSchema, ac as SummaryConfig, ad as SummaryConfigSchema, ae as SynthesisEvent, af as SynthesisOptions, ag as SynthesisRequest, ah as SynthesisResult, ai as SynthesisStreamRequest, aj as SynthesisTextSegment, ak as TokenConfig, al as TokenConfigSchema, am as TranscriptionEvent, an as TranscriptionOptions, ao as TranscriptionRequest, ap as TranscriptionResult, aq as TranscriptionSegment, ar as TranscriptionStreamRequest, as as TranscriptionTimestampGranularity, at as TranscriptionWord, au as ensureAudioCredential, av as resolveAudioModel, aw as resolveImageModel, ax as resolveModelApi, ay as resolveModelEntry } from './ai-reasoning-types-CuITylkA.js';
+export { A as AIFunctions, a as AIInitOptions, b as AUDIO_WS_PATH, c as AudioFormatSchema, d as AudioWsClientMessage, e as AudioWsClientMessageSchema, f as AudioWsDoneMessage, g as AudioWsDoneMessageSchema, h as AudioWsEndMessage, i as AudioWsErrorMessage, j as AudioWsSegmentDoneMessage, k as AudioWsSegmentStartedMessage, l as AudioWsServerMessage, m as AudioWsSpeechMessage, n as AudioWsStartMessage, o as AudioWsStartMessageSchema, p as AudioWsTextMessage, q as AudioWsTextMessageSchema, r as AudioWsTranscriptMessage, C as CompressionStrategy, s as CompressionStrategySchema, G as GenerateImageRequest, t as GenerateImageResult, u as GeneratedImage, H as HaiAIError, I as ImageOperations, v as ImageSize, R as ReferenceImage, w as serializePlayableAudio } from './ai-audio-ws-protocol-COIRfUFF.js';
 export { A2AClientOperations, AIApiAdapter, AIClient, AIClientConfig, AudioClientConfig, AudioClientOperations, AudioTicketRequest, StreamOptions, StreamProgress, collectStreamContent, createA2AClient, createAIClient, createAudioClient, createUnconfiguredAudioClient, parseSSE } from './client/index.js';
 import '@a2a-js/sdk/server';
 import '@h-ai/core';

package/dist/browser.js

@@ -1,4 +1,4 @@
-export { A2AConfigSchema, A2ASkillConfigSchema, AIConfigSchema, AUDIO_WS_PATH, ApiTypeSchema, AudioConfigSchema, AudioFormatSchema, AudioModelEntrySchema, AudioProviderSchema, AudioWsClientMessageSchema, AudioWsDoneMessageSchema, AudioWsStartMessageSchema, AudioWsTextMessageSchema, CompressConfigSchema, CompressionStrategySchema, EmbeddingConfigSchema, EntityTypeSchema, FileConfigSchema, ImageConfigSchema, ImageModelEntrySchema, ImageProviderSchema, KnowledgeConfigSchema, LLMConfigSchema, MCPConfigSchema, MCPServerCapabilitiesSchema, MCPServerConfigSchema, MemoryConfigSchema, MemoryTypeSchema, ModelEntrySchema, ModelScenarioSchema, OptionalSecretSchema, RetrievalConfigSchema, RetrievalSourceSchema, SummaryConfigSchema, TokenConfigSchema, resolveAudioModel, resolveImageModel, resolveModelApi, resolveModelEntry, serializePlayableAudio } from './chunk-4FDOGXIS.js';
+export { A2AConfigSchema, A2ASkillConfigSchema, AIConfigSchema, AUDIO_WS_PATH, ApiTypeSchema, AudioConfigSchema, AudioFormatSchema, AudioModelEntrySchema, AudioOperationSchema, AudioOperationsSchema, AudioProviderSchema, AudioWsClientMessageSchema, AudioWsDoneMessageSchema, AudioWsStartMessageSchema, AudioWsTextMessageSchema, CompressConfigSchema, CompressionStrategySchema, EmbeddingConfigSchema, EntityTypeSchema, FileConfigSchema, ImageConfigSchema, ImageModelEntrySchema, ImageProviderSchema, KnowledgeConfigSchema, LLMConfigSchema, MCPConfigSchema, MCPServerCapabilitiesSchema, MCPServerConfigSchema, MemoryConfigSchema, MemoryTypeSchema, ModelEntrySchema, ModelScenarioSchema, OptionalSecretSchema, RetrievalConfigSchema, RetrievalSourceSchema, SummaryConfigSchema, TokenConfigSchema, ensureAudioCredential, resolveAudioModel, resolveImageModel, resolveModelApi, resolveModelEntry, serializePlayableAudio } from './chunk-PLNY4UUT.js';
 export { collectStreamContent, createA2AClient, createAIClient, createAudioClient, createUnconfiguredAudioClient, parseSSE } from './chunk-6RWS5OWJ.js';
 export { HaiAIError } from './chunk-JC7QBT3C.js';
 //# sourceMappingURL=browser.js.map

package/dist/{chunk-4FDOGXIS.js → chunk-PLNY4UUT.js}

@@ -83,6 +83,7 @@ var en_US_default = {
   ai_audioInvalidRequest: "Invalid audio request: {reason}",
   ai_audioModelNotFound: "Audio model '{model}' not found. Register it in audio.models of ai.init().",
   ai_audioMissingApiKey: "Audio provider '{provider}' is missing credentials (API Key or App/Access Key).",
+  ai_audioMissingBaseUrl: "Audio provider '{provider}' is missing an endpoint. Configure baseUrl in the model entry.",
   ai_audioUnsupportedInput: "Provider '{provider}' does not support this input mode: {reason}",
   ai_audioUpstreamError: "Audio upstream service error: {error}",
   ai_audioProtocolError: "Audio protocol error: {error}",
@@ -183,6 +184,7 @@ var zh_CN_default = {
   ai_audioInvalidRequest: "\u8BED\u97F3\u8BF7\u6C42\u53C2\u6570\u65E0\u6548\uFF1A{reason}",
   ai_audioModelNotFound: "\u8BED\u97F3\u6A21\u578B '{model}' \u672A\u627E\u5230\uFF0C\u8BF7\u5728 ai.init() \u7684 audio.models \u4E2D\u6CE8\u518C",
   ai_audioMissingApiKey: "\u8BED\u97F3\u5E73\u53F0 '{provider}' \u7F3A\u5C11\u51ED\u636E\uFF08API Key \u6216 App/Access Key\uFF09",
+  ai_audioMissingBaseUrl: "\u8BED\u97F3\u5E73\u53F0 '{provider}' \u7F3A\u5C11\u7AEF\u70B9\u5730\u5740\uFF0C\u8BF7\u5728\u6A21\u578B\u6761\u76EE\u4E2D\u914D\u7F6E baseUrl",
   ai_audioUnsupportedInput: "\u5E73\u53F0 '{provider}' \u4E0D\u652F\u6301\u6B64\u8F93\u5165\u65B9\u5F0F\uFF1A{reason}",
   ai_audioUpstreamError: "\u8BED\u97F3\u4E0A\u6E38\u670D\u52A1\u9519\u8BEF\uFF1A{error}",
   ai_audioProtocolError: "\u8BED\u97F3\u534F\u8BAE\u9519\u8BEF\uFF1A{error}",
@@ -475,7 +477,11 @@ var A2AConfigSchema = z.object({
     }).optional()
   }).optional()
 });
-var AudioProviderSchema = z.enum(["openai", "mimo", "qwen", "doubao"]);
+var AudioProviderSchema = z.enum(["openai", "mimo", "qwen", "doubao", "whisper", "indextts"]);
+var AudioOperationSchema = z.enum(["transcribe", "synthesize"]);
+var AudioOperationsSchema = z.array(AudioOperationSchema).min(1).refine((operations) => new Set(operations).size === operations.length, {
+  message: "audio model operations must be unique"
+});
 var AudioModelEntrySchema = z.object({
   /** 模型唯一标识（用于场景解析与请求显式指定） */
   id: z.string(),
@@ -484,14 +490,10 @@ var AudioModelEntrySchema = z.object({
   /** 厂商模型名（传给厂商 API 的实际模型名） */
   model: z.string(),
   /** 模型允许执行的操作；解析模型时会在调用厂商前校验 */
-  operations: z.union([
-    z.tuple([z.literal("transcribe")]),
-    z.tuple([z.literal("synthesize")]),
-    z.tuple([z.literal("transcribe"), z.literal("synthesize")])
-  ]),
+  operations: AudioOperationsSchema,
   /** API Key 覆盖（未提供时可按 Audio 配置继承 LLM 密钥，最后回退对应平台环境变量） */
   apiKey: OptionalSecretSchema,
-  /** HTTP / WebSocket 端点覆盖（未提供时使用平台默认端点） */
+  /** HTTP / WebSocket 端点覆盖（`whisper` / `indextts` 无默认端点，必须显式配置） */
   baseUrl: z.string().optional(),
   /** 火山引擎 App Key（`X-Api-App-Key`，旧版控制台 ASR 需要） */
   appKey: OptionalSecretSchema,
@@ -524,6 +526,7 @@ var AUDIO_PROVIDER_DEFAULT_BASE_URL = {
   qwen: "wss://dashscope.aliyuncs.com/api-ws/v1/realtime",
   doubao: "wss://openspeech.bytedance.com"
 };
+var AUDIO_PROVIDER_CREDENTIAL_OPTIONAL = /* @__PURE__ */ new Set(["whisper", "indextts"]);
 function doubaoDefaultResourceId(operation) {
   return operation === "transcribe" ? "volc.bigasr.sauc.duration" : "seed-tts-2.0";
 }
@@ -537,6 +540,10 @@ function audioProviderEnvApiKey(provider) {
       return process.env.HAI_AI_AUDIO_QWEN_API_KEY ?? process.env.DASHSCOPE_API_KEY;
     case "doubao":
       return process.env.HAI_AI_AUDIO_DOUBAO_API_KEY ?? process.env.VOLC_API_KEY;
+    case "whisper":
+      return process.env.HAI_AI_AUDIO_WHISPER_API_KEY;
+    case "indextts":
+      return process.env.HAI_AI_AUDIO_INDEXTTS_API_KEY;
   }
 }
 function resolveAudioModel(audioConfig, operation, explicit, llmApiKey) {
@@ -554,15 +561,15 @@ function resolveAudioModel(audioConfig, operation, explicit, llmApiKey) {
     );
   }
   const apiKey = entry.apiKey ?? (audioConfig.inheritLlmApiKey ? llmApiKey : void 0) ?? audioProviderEnvApiKey(entry.provider);
-  const hasDoubaoLegacy = Boolean(entry.appKey && entry.accessKey);
-  if (!apiKey && !(entry.provider === "doubao" && hasDoubaoLegacy))
-    return err(HaiAIError.CONFIGURATION_ERROR, aiM("ai_audioMissingApiKey", { params: { provider: entry.provider } }));
+  const baseUrl = entry.baseUrl ?? AUDIO_PROVIDER_DEFAULT_BASE_URL[entry.provider];
+  if (!baseUrl)
+    return err(HaiAIError.CONFIGURATION_ERROR, aiM("ai_audioMissingBaseUrl", { params: { provider: entry.provider } }));
   return ok({
     id: entry.id,
     provider: entry.provider,
     model: entry.model,
     apiKey,
-    baseUrl: entry.baseUrl ?? AUDIO_PROVIDER_DEFAULT_BASE_URL[entry.provider],
+    baseUrl,
     appKey: entry.appKey ?? process.env.VOLC_APP_KEY,
     accessKey: entry.accessKey ?? process.env.VOLC_ACCESS_KEY,
     resourceId: entry.resourceId ?? (entry.provider === "doubao" ? doubaoDefaultResourceId(operation) : ""),
@@ -570,6 +577,14 @@ function resolveAudioModel(audioConfig, operation, explicit, llmApiKey) {
     timeout: entry.timeout ?? 6e4
   });
 }
+function ensureAudioCredential(model) {
+  if (AUDIO_PROVIDER_CREDENTIAL_OPTIONAL.has(model.provider))
+    return null;
+  const hasDoubaoLegacy = model.provider === "doubao" && Boolean(model.appKey && model.accessKey);
+  if (model.apiKey || hasDoubaoLegacy)
+    return null;
+  return core.error.buildHaiErrorInst(HaiAIError.CONFIGURATION_ERROR, aiM("ai_audioMissingApiKey", { params: { provider: model.provider } }));
+}
 var ImageProviderSchema = z.enum(["openai", "google", "qwen", "seedream", "pollinations"]);
 var ImageModelEntrySchema = z.object({
   /** 模型唯一标识（供请求选择） */
@@ -716,6 +731,6 @@ var AudioWsTextMessageSchema = AudioWsTextMessageSchema$1;
 var AudioWsDoneMessageSchema = AudioWsDoneMessageSchema$1;
 var AudioWsClientMessageSchema = AudioWsClientMessageSchema$1;
 
-export { A2AConfigSchema, A2ASkillConfigSchema, AIConfigSchema, AUDIO_WS_PATH, ApiTypeSchema, AudioConfigSchema, AudioFormatSchema, AudioModelEntrySchema, AudioProviderSchema, AudioWsClientMessageSchema, AudioWsDoneMessageSchema, AudioWsStartMessageSchema, AudioWsTextMessageSchema, CompressConfigSchema, CompressionStrategySchema, EmbeddingConfigSchema, EntityTypeSchema, FileConfigSchema, ImageConfigSchema, ImageModelEntrySchema, ImageProviderSchema, KnowledgeConfigSchema, LLMConfigSchema, MCPConfigSchema, MCPServerCapabilitiesSchema, MCPServerConfigSchema, MemoryConfigSchema, MemoryTypeSchema, ModelEntrySchema, ModelScenarioSchema, OptionalSecretSchema, RetrievalConfigSchema, RetrievalSourceSchema, SummaryConfigSchema, TokenConfigSchema, aiM, resolveAudioModel, resolveImageModel, resolveModelApi, resolveModelEntry, serializePlayableAudio };
-//# sourceMappingURL=chunk-4FDOGXIS.js.map
-//# sourceMappingURL=chunk-4FDOGXIS.js.map
+export { A2AConfigSchema, A2ASkillConfigSchema, AIConfigSchema, AUDIO_WS_PATH, ApiTypeSchema, AudioConfigSchema, AudioFormatSchema, AudioModelEntrySchema, AudioOperationSchema, AudioOperationsSchema, AudioProviderSchema, AudioWsClientMessageSchema, AudioWsDoneMessageSchema, AudioWsStartMessageSchema, AudioWsTextMessageSchema, CompressConfigSchema, CompressionStrategySchema, EmbeddingConfigSchema, EntityTypeSchema, FileConfigSchema, ImageConfigSchema, ImageModelEntrySchema, ImageProviderSchema, KnowledgeConfigSchema, LLMConfigSchema, MCPConfigSchema, MCPServerCapabilitiesSchema, MCPServerConfigSchema, MemoryConfigSchema, MemoryTypeSchema, ModelEntrySchema, ModelScenarioSchema, OptionalSecretSchema, RetrievalConfigSchema, RetrievalSourceSchema, SummaryConfigSchema, TokenConfigSchema, aiM, ensureAudioCredential, resolveAudioModel, resolveImageModel, resolveModelApi, resolveModelEntry, serializePlayableAudio };
+//# sourceMappingURL=chunk-PLNY4UUT.js.map
+//# sourceMappingURL=chunk-PLNY4UUT.js.map