@henjicc/ai-sdk 0.8.0 → 0.10.0

package/dist/providers/bailian.js

@@ -1,26 +1,60 @@
 import { AiRuntimeError } from '../runtime/AiRuntimeError.js';
+import { POLL_QUERY_FAILED, pollUntilResult } from '../protocols/polling.js';
 import { getPointer, isJsonObject, normalizeEndpoint, pushUniqueUrl, readJsonResponse, stringAt } from './helpers.js';
 import { fetchProvider } from './provider-fetch.js';
 import { hexAudioToDataUri } from './audio-data.js';
 const BAILIAN_BASE_URL = 'https://dashscope.aliyuncs.com';
+/** 只支持异步调用的 DashScope 路由：提交必须带 `X-DashScope-Async: enable`，结果按 `/api/v1/tasks/{task_id}` 查询。 */
+const ASYNC_ROUTES = new Set(['/api/v1/services/aigc/video-generation/video-synthesis']);
+function isAsyncRoute(route) {
+    return ASYNC_ROUTES.has(route.replace(/\/+$/, ''));
+}
+async function readBailianJson(response, model) {
+    return await readJsonResponse(response, 'Bailian', (body) => {
+        const message = ['/message', '/error/message', '/output/base_resp/status_msg']
+            .map((pointer) => stringAt(body, pointer))
+            .find((value) => value && /product is not activated|model.*(?:not activated|not authorized)|产品未开通|模型未开通/i.test(value));
+        if (message) {
+            throw new AiRuntimeError('provider_product_not_activated', `百炼账号未开通该模型（${model ?? '当前模型'}），请到百炼控制台 https://bailian.console.aliyun.com/ 开通后重试；开通前不要重复提交。`, {
+                retryable: false, category: 'authentication', status: response.status,
+                providerCode: stringAt(body, '/code') ?? stringAt(body, '/error/code') ?? getPointer(body, '/output/base_resp/status_code'),
+                providerMessage: message,
+            });
+        }
+    });
+}
 export async function execute(input) {
+    const asyncTask = isAsyncRoute(input.route);
+    const body = JSON.stringify(input.body);
     const response = await fetchProvider('Bailian', normalizeEndpoint(BAILIAN_BASE_URL, input.route), {
         method: input.method,
         headers: {
             Authorization: `Bearer ${input.apiKey}`,
             'Content-Type': 'application/json',
+            ...(asyncTask ? { 'X-DashScope-Async': 'enable' } : {}),
+            // 本地素材经临时存储上传成 oss:// 后，官方要求带这个头才会解析。
+            ...(body.includes('"oss://') ? { 'X-DashScope-OssResourceResolve': 'enable' } : {}),
         },
-        body: JSON.stringify(input.body),
+        body,
         signal: input.signal,
     }, {
         transport: input.runtime.transport,
         retryPreconnectOnce: true,
     });
-    const payload = await readJsonResponse(response, 'Bailian');
+    const payload = await readBailianJson(response, stringAt(input.body, '/model'));
     const errorCode = stringAt(payload, '/code');
     if (errorCode) {
         throw new AiRuntimeError('provider_task_failed', stringAt(payload, '/message') ?? errorCode);
     }
+    if (asyncTask) {
+        const taskStatus = stringAt(payload, '/output/task_status');
+        if (taskStatus === 'FAILED' || taskStatus === 'CANCELED')
+            throw taskFailure(payload, taskStatus);
+        const taskId = stringAt(payload, '/output/task_id')?.trim();
+        if (!taskId)
+            throw new AiRuntimeError('invalid_response', 'Bailian async task response has no task_id');
+        return { status: 'pending', url: '', taskId, metadata: payload };
+    }
     const speechStatus = getPointer(payload, '/output/base_resp/status_code');
     if (typeof speechStatus === 'number' && speechStatus !== 0) {
         throw new AiRuntimeError('provider_task_failed', stringAt(payload, '/output/base_resp/status_msg') ?? `MiniMax status ${speechStatus}`);
@@ -54,8 +88,41 @@ export async function execute(input) {
     }
     return { status: 'completed', url: urls.join('|||'), metadata: payload };
 }
-export async function continuePolling(_input) {
-    throw new AiRuntimeError('unsupported_provider', 'Bailian official image models use synchronous generation');
+export async function continuePolling(input) {
+    if (!isAsyncRoute(input.route)) {
+        throw new AiRuntimeError('unsupported_provider', 'Bailian official image models use synchronous generation');
+    }
+    const payload = await pollUntilResult(input, async () => {
+        const response = await fetchProvider('Bailian', `${BAILIAN_BASE_URL}/api/v1/tasks/${encodeURIComponent(input.taskId)}`, {
+            headers: { Authorization: `Bearer ${input.apiKey}` },
+            signal: input.signal,
+        }, {
+            transport: input.runtime.transport,
+            retryPreconnectOnce: true,
+        });
+        const result = await readBailianJson(response, undefined);
+        const status = stringAt(result, '/output/task_status');
+        if (status === 'SUCCEEDED')
+            return result;
+        if (status === 'PENDING' || status === 'RUNNING')
+            return undefined;
+        if (status === 'FAILED' || status === 'CANCELED')
+            throw taskFailure(result, status);
+        // UNKNOWN 表示任务不存在或已过 24 小时查询期；与缺状态一样计入连续查询失败，不立刻判死。
+        return POLL_QUERY_FAILED;
+    });
+    const videoUrl = stringAt(payload, '/output/video_url');
+    if (!videoUrl || !/^https?:\/\//i.test(videoUrl)) {
+        throw new AiRuntimeError('empty_result', `Bailian task has no video_url (task_id=${input.taskId})`);
+    }
+    return { status: 'completed', url: videoUrl, taskId: input.taskId, metadata: payload };
+}
+function taskFailure(payload, status) {
+    const code = stringAt(payload, '/output/code');
+    const message = stringAt(payload, '/output/message') ?? (status === 'CANCELED' ? '任务已取消' : 'Bailian task failed');
+    return new AiRuntimeError('provider_task_failed', code ? `${code}: ${message}` : message, {
+        providerCode: code, providerMessage: message,
+    });
 }
 function extractImageUrls(value) {
     const urls = [];

package/dist/providers/fal.js

@@ -121,9 +121,19 @@ function requireFalQueueUrl(value, field) {
     }
     return parsed.toString();
 }
+/**
+ * `draft_cache` 是 FLUX 3 草稿端点返回的加密缓存包，只用于转正片，不是生成结果；
+ * 收进媒体地址会被宿主当成第二段视频下载。它仍原样留在 metadata 里供续接读取。
+ */
 function extractUrls(payload) {
     const urls = [];
-    collectDeepUrls(payload, urls);
+    if (isJsonObject(payload) && 'draft_cache' in payload) {
+        const { draft_cache: _draftCache, ...media } = payload;
+        collectDeepUrls(media, urls);
+    }
+    else {
+        collectDeepUrls(payload, urls);
+    }
     return urls;
 }
 function enrichRequestId(value, requestId) {

package/dist/providers/helpers.d.ts

@@ -1,6 +1,6 @@
 import type { JsonObject, JsonValue } from '../types/runtime.js';
 export declare function normalizeEndpoint(base: string, route: string): string;
-export declare function readJsonResponse(response: Response, provider: string): Promise<JsonValue>;
+export declare function readJsonResponse(response: Response, provider: string, inspect?: (payload: JsonValue) => void): Promise<JsonValue>;
 export declare function asObject(value: JsonValue): JsonObject;
 export declare function getPointer(value: JsonValue, pointer: string): JsonValue | undefined;
 export declare function stringAt(value: JsonValue, pointer: string): string | undefined;

package/dist/providers/helpers.js

@@ -1,4 +1,5 @@
 import { AiRuntimeError } from '../runtime/AiRuntimeError.js';
+import { assertProviderAccess } from './providerAccessErrors.js';
 export function normalizeEndpoint(base, route) {
     if (route.startsWith('http://') || route.startsWith('https://')) {
         return route;
@@ -8,13 +9,24 @@ export function normalizeEndpoint(base, route) {
     }
     return `${base}/${route}`;
 }
-export async function readJsonResponse(response, provider) {
-    const payload = await response.json().catch((error) => {
+export async function readJsonResponse(response, provider, inspect) {
+    const text = await response.text();
+    let payload;
+    try {
+        payload = JSON.parse(text);
+    }
+    catch (error) {
+        if (!response.ok) {
+            assertProviderAccess(provider, {}, response.status);
+            throw new AiRuntimeError('provider_http_error', `${provider} HTTP ${response.status}: ${text.slice(0, 500)}`, { status: response.status });
+        }
         const message = error instanceof Error ? error.message : String(error);
-        throw new AiRuntimeError('invalid_json', message);
-    });
+        throw new AiRuntimeError('invalid_json', message, { status: response.status });
+    }
+    inspect?.(payload);
+    assertProviderAccess(provider, payload, response.status);
     if (!response.ok) {
-        throw new AiRuntimeError('provider_http_error', `${provider} HTTP ${response.status}: ${JSON.stringify(payload)}`);
+        throw new AiRuntimeError('provider_http_error', `${provider} HTTP ${response.status}: ${JSON.stringify(payload)}`, { status: response.status });
     }
     return payload;
 }

package/dist/providers/kie.js

@@ -42,6 +42,7 @@ async function sendCreateTask(input, endpoint) {
     const code = getPointer(payload, '/code');
     const success = getPointer(payload, '/success');
     if ((typeof code === 'number' && code !== 200) || success === false) {
+        assertKieTaskFailure(code, stringAt(payload, '/msg') ?? '');
         throw new AiRuntimeError('provider_task_failed', stringAt(payload, '/msg') ?? 'KIE create task failed');
     }
     return payload;
@@ -58,16 +59,18 @@ async function pollTask(input) {
         });
         const payload = await readJsonResponse(response, 'KIE');
         const code = getPointer(payload, '/code');
-        if (typeof code === 'number' && code !== 200) {
-            // 查询本身没成功（含 taskId 已被清理的情况），不是"任务还在跑"
-            return POLL_QUERY_FAILED;
+        if ((typeof code === 'number' && code !== 200) || getPointer(payload, '/success') === false) {
+            // 查询失败保留供应商事实；鉴权/余额错误已由 readJsonResponse 立即终止。
+            const message = stringAt(payload, '/msg') ?? stringAt(payload, '/message') ?? 'provider reported a non-success status';
+            throw new AiRuntimeError('provider_http_error', `KIE recordInfo code=${code ?? 'unknown'}: ${message}`, {
+                status: response.status, providerCode: code, providerMessage: message,
+            });
         }
-        if (getPointer(payload, '/success') === false)
-            return POLL_QUERY_FAILED;
         const state = (stringAt(payload, '/data/state') ?? '').trim().toLowerCase();
         if (state === 'success')
             return payload;
         if (state === 'fail') {
+            assertKieTaskFailure(getPointer(payload, '/data/failCode'), stringAt(payload, '/data/failMsg') ?? '');
             throw new AiRuntimeError('provider_task_failed', stringAt(payload, '/data/failMsg') ?? 'KIE task failed');
         }
         if (['waiting', 'queuing', 'generating'].includes(state))
@@ -75,6 +78,14 @@ async function pollTask(input) {
         return POLL_QUERY_FAILED;
     });
 }
+function assertKieTaskFailure(code, message) {
+    if (/^Internal Error(?:,|\.|$)/i.test(message.trim())) {
+        throw new AiRuntimeError('provider_internal_error', 'KIE 服务内部错误，可稍后手动重试。', { retryable: true, terminal: true, providerCode: code, providerMessage: message });
+    }
+    if (String(code) === '422') {
+        throw new AiRuntimeError('provider_invalid_request', `KIE 拒绝了请求参数，请修改后重试：${message}`, { retryable: false, providerCode: code, providerMessage: message });
+    }
+}
 function extractTaskId(payload) {
     return stringAt(payload, '/data/taskId');
 }

package/dist/providers/metadata.js

@@ -82,8 +82,8 @@ export const PROVIDER_METADATA = [
     {
         providerId: 'minimax',
         displayName: 'MiniMax',
-        websiteUrl: 'https://platform.minimaxi.com/',
-        apiKeyUrl: 'https://platform.minimaxi.com/console/access?tab=api-keys',
+        websiteUrl: 'https://platform.minimax.cn/',
+        apiKeyUrl: 'https://platform.minimax.cn/console/access?tab=api-keys',
         websiteUrlKind: 'official',
         docs: 'docs/llm-adaptation/供应商/MiniMax.md',
     },

package/dist/providers/openai-speech.d.ts

@@ -0,0 +1,6 @@
+import type { ProviderExecutionInput, ProviderExecutionResult } from '../types/runtime.js';
+/**
+ * OpenAI 兼容 `/audio/speech`：同步返回音频二进制，错误时返回 JSON 错误体。
+ * 返回的音频转成 data URI，由宿主按普通媒体结果落盘。
+ */
+export declare function executeOpenAiCompatibleSpeech(providerName: string, baseUrl: string, input: ProviderExecutionInput, defaultFormat: string): Promise<ProviderExecutionResult>;

package/dist/providers/openai-speech.js

@@ -0,0 +1,31 @@
+import { audioBytesToDataUri } from './audio-data.js';
+import { isJsonObject, normalizeEndpoint, readJsonResponse } from './helpers.js';
+import { fetchProvider } from './provider-fetch.js';
+/**
+ * OpenAI 兼容 `/audio/speech`：同步返回音频二进制，错误时返回 JSON 错误体。
+ * 返回的音频转成 data URI，由宿主按普通媒体结果落盘。
+ */
+export async function executeOpenAiCompatibleSpeech(providerName, baseUrl, input, defaultFormat) {
+    const response = await fetchProvider(providerName, normalizeEndpoint(baseUrl, input.route), {
+        method: 'POST',
+        headers: {
+            Authorization: `Bearer ${input.apiKey}`,
+            'Content-Type': 'application/json',
+        },
+        body: JSON.stringify(input.body),
+        signal: input.signal,
+    }, {
+        transport: input.runtime.transport,
+        retryPreconnectOnce: true,
+    });
+    if (!response.ok) {
+        await readJsonResponse(response, providerName);
+    }
+    const bytes = new Uint8Array(await response.arrayBuffer());
+    const format = isJsonObject(input.body) && typeof input.body.response_format === 'string' ? input.body.response_format : defaultFormat;
+    return {
+        status: 'completed',
+        url: audioBytesToDataUri(bytes, format),
+        metadata: { contentType: response.headers.get('content-type'), byteLength: bytes.byteLength },
+    };
+}

package/dist/providers/providerAccessErrors.d.ts

@@ -0,0 +1,3 @@
+import type { JsonValue } from '../types/runtime.js';
+/** 仅按供应商官方错误契约识别；未知 403/500 保留原分类，禁止猜测余额或模型开通状态。 */
+export declare function assertProviderAccess(provider: string, payload: JsonValue, status: number): void;

package/dist/providers/providerAccessErrors.js

@@ -0,0 +1,73 @@
+import { AiRuntimeError } from '../runtime/AiRuntimeError.js';
+const object = (value) => value && typeof value === 'object' && !Array.isArray(value) ? value : {};
+/** 仅按供应商官方错误契约识别；未知 403/500 保留原分类，禁止猜测余额或模型开通状态。 */
+export function assertProviderAccess(provider, payload, status) {
+    const body = object(payload);
+    const error = object(body.error ?? body.errors);
+    const code = String(error.code ?? body.code ?? '');
+    const providerMessage = String(error.message ?? body.message ?? body.msg ?? '');
+    if (provider === 'KIE' && (['401', '403'].includes(code) || status === 401 || status === 403)) {
+        throw new AiRuntimeError('provider_authentication_failed', 'KIE 鉴权失败，请检查 API Key 与账号权限。控制台：https://kie.ai/api-key', {
+            retryable: false, category: 'authentication', providerCode: code || String(status), status,
+            providerMessage: String(error.message ?? body.message ?? body.msg ?? ''),
+        });
+    }
+    let kind;
+    let accountNotVerified = false;
+    if (provider === 'Bailian') {
+        if (code === 'Arrearage')
+            kind = 'insufficient_balance';
+        else if (code === 'AccessDenied.Unpurchased')
+            kind = 'product_not_activated';
+        else if (code === 'Model.AccessDenied')
+            kind = 'model_access_denied';
+    }
+    else if (provider === 'KIE' && (code === '402' || status === 402))
+        kind = 'insufficient_balance';
+    else if (provider === 'PPIO') {
+        if (['BILLING_BALANCE_NOT_ENOUGH', 'NOT_ENOUGH_BALANCE'].includes(code))
+            kind = 'insufficient_balance';
+        else if (['API_NOT_ALLOWED', 'ACCESS_DENY'].includes(code))
+            kind = 'model_access_denied';
+    }
+    else if (provider === 'SiliconFlow') {
+        if (status === 402)
+            kind = 'insufficient_balance';
+        else if (status === 403)
+            kind = 'model_access_denied';
+    }
+    else if (provider === 'APIMart') {
+        if (code === '402' || status === 402)
+            kind = 'insufficient_balance';
+        else if (code === '403' || status === 403)
+            kind = 'model_access_denied';
+    }
+    else if (provider === 'ModelScope') {
+        // 魔搭 API 推理只对完成邮箱验证的账号开放；这个 403 只有英文 message，没有错误码。
+        if (status === 403 && /verified accounts|email verification/i.test(providerMessage)) {
+            kind = 'product_not_activated';
+            accountNotVerified = true;
+        }
+    }
+    else if (provider === 'Volcengine Ark') {
+        if (code === 'AccountOverdueError')
+            kind = 'insufficient_balance';
+        else if (['ModelNotOpen', 'OperationDenied.ServiceNotOpen'].includes(code))
+            kind = 'product_not_activated';
+        else if (['AccessDenied', 'OperationDenied.PermissionDenied'].includes(code))
+            kind = 'model_access_denied';
+    }
+    if (!kind)
+        return;
+    const name = provider === 'Bailian' ? '百炼' : provider === 'SiliconFlow' ? '硅基流动' : provider === 'Volcengine Ark' ? '火山方舟' : provider === 'ModelScope' ? '魔搭' : provider;
+    const destination = { Bailian: 'https://bailian.console.aliyun.com/', KIE: 'https://kie.ai/', PPIO: 'https://ppio.com/console', SiliconFlow: 'https://cloud.siliconflow.cn/', APIMart: 'https://apimart.ai/', 'Volcengine Ark': 'https://console.volcengine.com/ark/', ModelScope: 'https://modelscope.cn/my/settings/account' };
+    const action = accountNotVerified ? '账号需先完成邮箱验证才能调用 API 推理，请到账号设置完成验证后重试'
+        : kind === 'insufficient_balance' ? '账户余额不足或已欠费，请充值后重试'
+            : kind === 'product_not_activated' ? '账号未开通该模型，请先开通模型服务'
+                : '账号没有该模型的调用权限，请检查模型授权和实名认证';
+    throw new AiRuntimeError(`provider_${kind}`, `${name}${action}。控制台：${destination[provider]}`, {
+        retryable: false, category: 'authentication', providerCode: code, status,
+        providerMessage,
+        scope: kind === 'insufficient_balance' || accountNotVerified ? 'account' : 'model',
+    });
+}

package/dist/providers/siliconflow.js

@@ -1,33 +1,8 @@
 import { AiRuntimeError } from '../runtime/AiRuntimeError.js';
-import { audioBytesToDataUri } from './audio-data.js';
-import { normalizeEndpoint } from './helpers.js';
-import { isJsonObject } from './helpers.js';
-import { fetchProvider } from './provider-fetch.js';
+import { executeOpenAiCompatibleSpeech } from './openai-speech.js';
 const BASE_URL = 'https://api.siliconflow.cn';
 export async function execute(input) {
-    const response = await fetchProvider('SiliconFlow', normalizeEndpoint(BASE_URL, input.route), {
-        method: 'POST',
-        headers: {
-            Authorization: `Bearer ${input.apiKey}`,
-            'Content-Type': 'application/json',
-        },
-        body: JSON.stringify(input.body),
-        signal: input.signal,
-    }, {
-        transport: input.runtime.transport,
-        retryPreconnectOnce: true,
-    });
-    if (!response.ok) {
-        await response.body?.cancel().catch(() => undefined);
-        throw new AiRuntimeError('provider_task_failed', `SiliconFlow speech request failed (HTTP ${response.status})`);
-    }
-    const bytes = new Uint8Array(await response.arrayBuffer());
-    const requestedFormat = isJsonObject(input.body) && typeof input.body.response_format === 'string' ? input.body.response_format : 'mp3';
-    const url = audioBytesToDataUri(bytes, requestedFormat);
-    return {
-        status: 'completed', url,
-        metadata: { contentType: response.headers.get('content-type'), byteLength: bytes.byteLength },
-    };
+    return executeOpenAiCompatibleSpeech('SiliconFlow', BASE_URL, input, 'mp3');
 }
 export async function continuePolling(_input) {
     throw new AiRuntimeError('unsupported_provider', 'SiliconFlow speech response is synchronous');

package/dist/providers/volcengine-speech.d.ts

@@ -1,3 +1,3 @@
 import type { ProviderContinuePollingInput, ProviderExecutionInput, ProviderExecutionResult } from '../types/runtime.js';
 export declare function execute(input: ProviderExecutionInput): Promise<ProviderExecutionResult>;
-export declare function continuePolling(_input: ProviderContinuePollingInput): Promise<ProviderExecutionResult>;
+export declare function continuePolling(input: ProviderContinuePollingInput): Promise<ProviderExecutionResult>;

package/dist/providers/volcengine-speech.js

@@ -5,6 +5,7 @@ import { fromBase64 } from '../upload/base64.js';
 import { audioBytesToDataUri } from './audio-data.js';
 import { isJsonObject, normalizeEndpoint } from './helpers.js';
 import { fetchProvider } from './provider-fetch.js';
+import { executeVoiceClone, pollVoiceClone } from './volcengine-voice-clone.js';
 const BASE_URL = 'https://openspeech.bytedance.com';
 const ENDPOINT = '/api/v3/tts/unidirectional';
 function decodeChunk(encoded) {
@@ -15,7 +16,7 @@ function decodeChunk(encoded) {
         throw new AiRuntimeError('invalid_response', 'Volcengine speech returned invalid base64 audio');
     }
 }
-function parseFrame(payload, audio, usage) {
+function parseFrame(payload, audio, usage, items) {
     if (!payload)
         return;
     let value;
@@ -32,8 +33,19 @@ function parseFrame(payload, audio, usage) {
         audio.push(decodeChunk(value.data));
     if (isJsonObject(value.usage) && typeof value.usage.text_words === 'number')
         usage.textWords = value.usage.text_words;
+    if (isJsonObject(value.sentence) && Array.isArray(value.sentence.words)) {
+        for (const word of value.sentence.words) {
+            if (!isJsonObject(word) || typeof word.word !== 'string' || typeof word.startTime !== 'number' || typeof word.endTime !== 'number'
+                || !Number.isFinite(word.startTime) || !Number.isFinite(word.endTime) || word.startTime < 0 || word.endTime < word.startTime) {
+                throw new AiRuntimeError('invalid_response', 'Volcengine speech returned invalid subtitle timing');
+            }
+            items.push({ text: word.word, startMs: word.startTime * 1000, endMs: word.endTime * 1000,
+                ...(typeof word.confidence === 'number' && word.confidence >= 0 && word.confidence <= 1 ? { confidence: word.confidence } : {}),
+            });
+        }
+    }
 }
-function consumeFrames(buffer, audio, usage) {
+function consumeFrames(buffer, audio, usage, items) {
     let start = -1;
     let depth = 0;
     let quoted = false;
@@ -67,7 +79,7 @@ function consumeFrames(buffer, audio, usage) {
         else if (character === '}') {
             depth -= 1;
             if (depth === 0) {
-                parseFrame(buffer.slice(start, index + 1), audio, usage);
+                parseFrame(buffer.slice(start, index + 1), audio, usage, items);
                 start = -1;
                 consumed = index + 1;
             }
@@ -76,6 +88,8 @@ function consumeFrames(buffer, audio, usage) {
     return buffer.slice(start < 0 ? consumed : start);
 }
 export async function execute(input) {
+    if (input.route === '/api/v3/tts/voice_clone')
+        return executeVoiceClone(input);
     if (!isJsonObject(input.body) || !isJsonObject(input.body.req_params)) {
         throw new AiRuntimeError('invalid_request', 'Volcengine speech requires req_params');
     }
@@ -105,22 +119,27 @@ export async function execute(input) {
     const decoder = createUtf8StreamDecoder();
     const audio = [];
     const usage = {};
+    const items = [];
     let pending = '';
     const cancelReader = () => { void reader.cancel(input.signal?.reason).catch(() => undefined); };
     input.signal?.addEventListener('abort', cancelReader, { once: true });
     try {
-        while (true) {
+        for (;;) {
             input.signal?.throwIfAborted();
             const { done, value } = await reader.read();
             input.signal?.throwIfAborted();
             pending += done ? decoder.decode() : decoder.decode(value, { stream: true });
-            pending = consumeFrames(pending, audio, usage);
+            pending = consumeFrames(pending, audio, usage, items);
             if (done)
                 break;
         }
         if (pending.trim())
             throw new AiRuntimeError('invalid_response', 'Volcengine speech returned incomplete JSON frame');
     }
+    catch (error) {
+        await reader.cancel(error).catch(() => undefined);
+        throw error;
+    }
     finally {
         input.signal?.removeEventListener('abort', cancelReader);
         reader.releaseLock();
@@ -135,9 +154,13 @@ export async function execute(input) {
     return {
         status: 'completed',
         url: audioBytesToDataUri(joined, 'mp3'),
-        metadata: { byteLength: total, ...(usage.textWords === undefined ? {} : { textWords: usage.textWords }) },
+        metadata: { byteLength: total, ...(usage.textWords === undefined ? {} : { textWords: usage.textWords }),
+            ...(items.length ? { outputTimings: [{ outputIndex: 0, timings: { granularity: 'char', items } }] } : {}),
+            ...(resourceId === 'seed-icl-2.0' && typeof input.body.req_params.speaker === 'string'
+                ? { activatedVoiceId: input.body.req_params.speaker } : {}),
+        },
     };
 }
-export async function continuePolling(_input) {
-    throw new AiRuntimeError('unsupported_provider', 'Volcengine speech uses one HTTP stream');
+export async function continuePolling(input) {
+    return pollVoiceClone(input);
 }

package/dist/providers/volcengine-voice-clone.d.ts

@@ -0,0 +1,3 @@
+import type { ProviderExecutionInput, ProviderContinuePollingInput, ProviderExecutionResult } from '../types/runtime.js';
+export declare function executeVoiceClone(input: ProviderExecutionInput): Promise<ProviderExecutionResult>;
+export declare function pollVoiceClone(input: ProviderContinuePollingInput): Promise<ProviderExecutionResult>;

package/dist/providers/volcengine-voice-clone.js

@@ -0,0 +1,138 @@
+import { AiRuntimeError } from '../runtime/AiRuntimeError.js';
+import { volcengineRequestId } from '../capabilities/speech-recognition/volcengine/request-id.js';
+import { pollUntilResult } from '../protocols/polling.js';
+import { parseDataUri } from '../upload/media-binary.js';
+import { toBase64 } from '../upload/base64.js';
+import { isJsonObject } from './helpers.js';
+import { fetchProvider } from './provider-fetch.js';
+const TASK_PREFIX = 'volc-voice:';
+const MAX_AUDIO_BYTES = 10_000_000;
+function taskId(identity) { return TASK_PREFIX + encodeURIComponent(JSON.stringify(identity)); }
+function decodeTask(value) {
+    try {
+        if (!value.startsWith(TASK_PREFIX))
+            throw new Error('prefix');
+        const data = JSON.parse(decodeURIComponent(value.slice(TASK_PREFIX.length)));
+        if (!isJsonObject(data) || typeof data.speaker !== 'string' || !data.speaker.trim()
+            || typeof data.name !== 'string' || !data.name.trim() || typeof data.postpaid !== 'boolean'
+            || typeof data.submission !== 'string' || !data.submission.trim())
+            throw new Error('identity');
+        return { speaker: data.speaker, name: data.name, postpaid: data.postpaid, submission: data.submission };
+    }
+    catch {
+        throw new AiRuntimeError('invalid_task_id', '无效的音色训练任务');
+    }
+}
+function speakerBody(identity) {
+    return identity.postpaid
+        ? { speaker_id: 'custom_speaker_id', custom_speaker_id: identity.speaker }
+        : { speaker_id: identity.speaker };
+}
+async function request(input, route, body) {
+    input.signal?.throwIfAborted();
+    const response = await fetchProvider('Volcengine Speech', `https://openspeech.bytedance.com${route}`, {
+        method: 'POST',
+        headers: { 'Content-Type': 'application/json', 'X-Api-Key': input.apiKey, 'X-Api-Request-Id': volcengineRequestId(input.requestId, undefined) },
+        body: JSON.stringify(body), signal: input.signal,
+    }, { transport: input.runtime.transport, retryPreconnectOnce: false });
+    const text = await response.text();
+    let payload;
+    try {
+        payload = JSON.parse(text);
+    }
+    catch {
+        throw new AiRuntimeError(response.ok ? 'invalid_json' : 'provider_http_error', `Volcengine Speech HTTP ${response.status}: 音色训练返回了非 JSON 响应`, { status: response.status });
+    }
+    if (!isJsonObject(payload))
+        throw new AiRuntimeError('invalid_response', '音色训练返回了无效响应');
+    if (!response.ok || (payload.code !== undefined && payload.code !== 0)) {
+        throw new AiRuntimeError('provider_task_failed', typeof payload.message === 'string' ? payload.message : `音色训练请求失败（HTTP ${response.status}）`, { status: response.status });
+    }
+    return payload;
+}
+function parseResult(payload, identity) {
+    if (typeof payload.speaker_id !== 'string' || !payload.speaker_id.trim()
+        || ![0, 1, 2, 3, 4].includes(payload.status)) {
+        throw new AiRuntimeError('invalid_response', '音色训练响应缺少有效音色或状态');
+    }
+    // 后付费响应可返回固定代号；实际自定义代号来自已提交任务身份。
+    if (payload.speaker_id !== identity.speaker && !(identity.postpaid && payload.speaker_id === 'custom_speaker_id')) {
+        throw new AiRuntimeError('invalid_response', '音色查询返回了其他音色');
+    }
+    if (payload.status === 0 || payload.status === 3) {
+        throw new AiRuntimeError(payload.status === 0 ? 'voice_not_found' : 'voice_training_failed', typeof payload.message === 'string' ? payload.message : payload.status === 0 ? '音色不存在或尚未训练' : '音色训练失败', { retryable: false });
+    }
+    const ready = payload.status === 2 || payload.status === 4;
+    const samples = Array.isArray(payload.speaker_status) ? payload.speaker_status : [];
+    const sample = samples.find(item => isJsonObject(item) && item.model_type === 5);
+    const demo = isJsonObject(sample) && typeof sample.demo_audio === 'string' ? sample.demo_audio : '';
+    if (demo && !/^https:\/\//i.test(demo))
+        throw new AiRuntimeError('invalid_response', '音色试听地址无效');
+    return {
+        status: ready ? 'completed' : 'pending', url: ready ? demo : '', taskId: taskId(identity),
+        metadata: {
+            ...payload,
+            clonedVoice: { id: identity.speaker, name: identity.name, status: ready ? 'ready' : 'training',
+                postpaid: identity.postpaid, activated: payload.status === 4,
+                ...(typeof payload.create_time === 'number' ? { createdAt: payload.create_time } : {}) },
+        },
+    };
+}
+export async function executeVoiceClone(input) {
+    if (!isJsonObject(input.body))
+        throw new AiRuntimeError('invalid_request', '音色克隆参数无效');
+    const body = input.body;
+    if (typeof body.clone_audio !== 'string' || typeof body.voice_name !== 'string' || !body.voice_name.trim()) {
+        throw new AiRuntimeError('invalid_request', '请填写音色名称并上传声音样本');
+    }
+    const existing = typeof body.speaker_id === 'string' ? body.speaker_id.trim() : '';
+    const custom = typeof body.custom_speaker_id === 'string' ? body.custom_speaker_id.trim() : '';
+    const identity = {
+        speaker: existing || custom || `henji${volcengineRequestId(input.requestId, undefined).replaceAll('-', '')}`,
+        name: body.voice_name.trim(), postpaid: !existing,
+        submission: volcengineRequestId(input.requestId, undefined),
+    };
+    const audio = parseDataUri(body.clone_audio) ?? await input.runtime.media.read(body.clone_audio);
+    if (!audio.bytes.length || audio.bytes.length > MAX_AUDIO_BYTES)
+        throw new AiRuntimeError('invalid_media', '声音样本必须非空且不超过 10 MB');
+    const formats = { 'audio/wav': 'wav', 'audio/x-wav': 'wav', 'audio/wave': 'wav', 'audio/mpeg': 'mp3', 'audio/mp3': 'mp3', 'audio/ogg': 'ogg', 'audio/mp4': 'm4a', 'audio/x-m4a': 'm4a', 'audio/aac': 'aac' };
+    const format = formats[audio.mimeType];
+    if (!format)
+        throw new AiRuntimeError('invalid_media', '请上传 WAV、MP3、OGG、M4A 或 AAC 音频');
+    const payload = {
+        ...speakerBody(identity), audio: { data: toBase64(audio.bytes), format },
+        language: body.language ?? 0, extra_params: body.extra_params ?? {},
+        ...(typeof body.text === 'string' && body.text.trim() ? { text: body.text } : {}),
+    };
+    // Existing slots are never overwritten blindly (activation makes training unavailable).
+    if (existing) {
+        const current = await request(input, '/api/v3/tts/get_voice', speakerBody(identity));
+        if (current.speaker_id !== existing || ![0, 1, 2, 3, 4].includes(current.status)) {
+            throw new AiRuntimeError('invalid_response', '无法确认已有音色的训练状态');
+        }
+        if (current.status === 4 || current.status === 1 || current.available_training_times === 0) {
+            throw new AiRuntimeError('invalid_request', '该音色已锁定、正在训练或没有剩余训练次数，请选择其他音色');
+        }
+    }
+    try {
+        return parseResult(await request(input, '/api/v3/tts/voice_clone', payload), identity);
+    }
+    catch (error) {
+        // A connection loss after sending is ambiguous. Keep a queryable identity;
+        // never send a second training request to guess whether the first succeeded.
+        if (error instanceof AiRuntimeError && error.code === 'provider_network_error' && error.details?.submissionState === 'unknown') {
+            return { status: 'pending', url: '', taskId: taskId(identity), metadata: {
+                    submissionUnconfirmed: true,
+                    clonedVoice: { id: identity.speaker, name: identity.name, status: 'training', postpaid: identity.postpaid, activated: false },
+                } };
+        }
+        throw error;
+    }
+}
+export async function pollVoiceClone(input) {
+    const identity = decodeTask(input.taskId);
+    return pollUntilResult(input, async () => {
+        const result = parseResult(await request(input, '/api/v3/tts/get_voice', speakerBody(identity)), identity);
+        return result.status === 'pending' ? undefined : result;
+    });
+}