@xopcai/xopc 0.0.241 → 0.0.243

package/dist/src/voice/realtime/speakable-segmenter.js

@@ -0,0 +1,87 @@
+import { stripMarkdown } from "../tts/preprocess.js";
+//#region src/voice/realtime/speakable-segmenter.ts
+const SENTENCE_END = /[。！？!?；;：:\n]/u;
+/** Streams spoken prose, keeping code and link targets out of synthesis. */
+var SpeakableSegmenter = class {
+	pending = "";
+	fence = null;
+	constructor(maxCharacters = 120) {
+		this.maxCharacters = maxCharacters;
+	}
+	push(delta) {
+		this.pending += delta;
+		const phrases = [];
+		while (this.pending) {
+			if (this.fence) {
+				const end = this.pending.indexOf(this.fence);
+				if (end < 0) {
+					this.pending = this.pending.slice(-2);
+					break;
+				}
+				this.pending = this.pending.slice(end + this.fence.length);
+				this.fence = null;
+				continue;
+			}
+			const openingFence = this.pending.match(/^\s*(```|~~~)/);
+			if (openingFence) {
+				this.pending = this.pending.trimStart();
+				const newline = this.pending.indexOf("\n");
+				if (newline < 0) break;
+				this.fence = openingFence[1];
+				this.pending = this.pending.slice(newline + 1);
+				continue;
+			}
+			if (/^\s*\|/.test(this.pending)) {
+				const newline = this.pending.indexOf("\n", this.pending.search(/\|/));
+				if (newline < 0) break;
+				this.pending = this.pending.slice(newline + 1);
+				continue;
+			}
+			let boundary = -1;
+			let brackets = 0;
+			let parentheses = 0;
+			let inlineCode = false;
+			for (let index = 0; index < this.pending.length; index += 1) {
+				if (!parentheses && /^https?:\/\//.test(this.pending.slice(index))) {
+					const end = this.pending.slice(index).search(/\s/);
+					if (end < 0) break;
+					index += end - 1;
+					continue;
+				}
+				const char = this.pending[index];
+				if (char === "`") inlineCode = !inlineCode;
+				if (char === "[") brackets += 1;
+				if (char === "]") brackets = Math.max(0, brackets - 1);
+				if (char === "(") parentheses += 1;
+				if (char === ")") parentheses = Math.max(0, parentheses - 1);
+				if (inlineCode || brackets || parentheses || char === "]" && this.pending[index + 1] === "(") continue;
+				if (SENTENCE_END.test(char)) {
+					boundary = index + 1;
+					break;
+				}
+				if (index + 1 >= this.maxCharacters) {
+					const window = this.pending.slice(0, index + 1);
+					const whitespace = Math.max(window.lastIndexOf(" "), window.lastIndexOf("，"), window.lastIndexOf(","));
+					boundary = whitespace > this.maxCharacters / 2 ? whitespace + 1 : index + 1;
+					break;
+				}
+			}
+			if (boundary < 0) break;
+			const phrase = this.toSpeech(this.pending.slice(0, boundary));
+			this.pending = this.pending.slice(boundary);
+			if (phrase) phrases.push(phrase);
+		}
+		return phrases;
+	}
+	flush() {
+		const phrase = this.fence || /^\s*(```|~~~)/.test(this.pending) ? "" : this.toSpeech(this.pending);
+		this.pending = "";
+		this.fence = null;
+		return phrase ? [phrase] : [];
+	}
+	toSpeech(text) {
+		return stripMarkdown(text).replace(/https?:\/\/\S+/g, "").replace(/[*_~]/g, "").replace(/^\s*\|.*\|\s*$/gm, "").trim();
+	}
+};
+//#endregion
+export { SpeakableSegmenter };

package/dist/src/voice/stt/list-providers.js

@@ -1,4 +1,4 @@
-import { listMediaUnderstandingProviders } from "../../media-understanding/registry.js";
+import { getMediaUnderstandingProvider, listMediaUnderstandingProviders } from "../../media-understanding/registry.js";
 import { resolveSTTProviderConfig } from "./factory.js";
 import { mergeSttConfigFromAppConfig } from "./config.js";
 import { getVoiceProviderMetadata } from "../metadata/registry.js";
@@ -35,6 +35,7 @@ function fallbackMetadata(providerId, aliases) {
 	};
 }
 function isSttProviderConfigured(providerId, config) {
+	if (!getMediaUnderstandingProvider(providerId)?.transcribeAudio) return false;
 	return resolveSTTProviderConfig(providerId, config) !== null;
 }
 /** List registered audio STT providers with configured state and UI metadata. */

package/dist/src/voice/stt/providers/alibaba-transcription.js

@@ -3,10 +3,12 @@ import { init_logger } from "../../../utils/logger.js";
 import { postJsonRequest } from "../../../media-shared/http/provider-http.js";
 import "../../../media-shared/http/index.js";
 import { registerMediaUnderstandingProvider } from "../../../media-understanding/registry.js";
+import { openDashScopeStreamingStt } from "../../dashscope/streaming-stt-session.js";
 //#region src/voice/stt/providers/alibaba-transcription.ts
 init_logger();
 const log = createLogger("STT:Alibaba");
 const DEFAULT_MODEL = "qwen-audio-3.0-asr-flash";
+const DEFAULT_STREAMING_MODEL = "qwen-audio-3.0-asr-flash-streaming";
 const DEFAULT_BASE_URL = "https://dashscope.aliyuncs.com/api/v1/services/aigc/multimodal-generation/generation";
 function audioFormat(request) {
 	const extension = request.fileName.split(".").pop()?.toLowerCase();
@@ -85,7 +87,17 @@ const alibabaTranscriptionProvider = {
 	envKey: "DASHSCOPE_API_KEY",
 	defaultModels: { audio: DEFAULT_MODEL },
 	autoPriority: { audio: 10 },
-	transcribeAudio
+	transcribeAudio,
+	streamingAudio: {
+		inputSampleRates: [16e3],
+		turnDetection: ["server_vad"],
+		defaultModel: DEFAULT_STREAMING_MODEL,
+		models: [DEFAULT_STREAMING_MODEL]
+	},
+	openAudioStream: (request) => openDashScopeStreamingStt({
+		...request,
+		model: request.model || DEFAULT_STREAMING_MODEL
+	})
 };
 registerMediaUnderstandingProvider(alibabaTranscriptionProvider);
 //#endregion

package/dist/src/voice/stt/providers/xopc-cloud-transcription.js

@@ -3,10 +3,12 @@ import { init_xopc_cloud_config, resolveXopcModelRouterUrl } from "../../../prov
 import { getProviderAuthService, init_provider_auth_service } from "../../../providers/provider-auth-service.js";
 import { compareCatalogModels } from "../../../providers/model-catalog-ranking.js";
 import { registerMediaUnderstandingProvider } from "../../../media-understanding/registry.js";
+import { openDashScopeStreamingStt } from "../../dashscope/streaming-stt-session.js";
 //#region src/voice/stt/providers/xopc-cloud-transcription.ts
 init_model_catalog_store();
 init_provider_auth_service();
 init_xopc_cloud_config();
+const DEFAULT_STREAMING_MODEL = "qwen-audio-3.0-asr-flash-streaming";
 function defaultModel() {
 	const source = getModelCatalogStore().getSource("xopc-cloud");
 	return source?.models.filter((model) => model.availability === "available" && model.kind === "stt").sort((left, right) => compareCatalogModels(left, right, source.recommended?.stt))[0]?.id;
@@ -44,6 +46,27 @@ const xopcCloudTranscriptionProvider = {
 			model,
 			...request.language ? { language: request.language } : {}
 		};
+	},
+	streamingAudio: {
+		inputSampleRates: [16e3],
+		turnDetection: ["server_vad"],
+		defaultModel: DEFAULT_STREAMING_MODEL,
+		models: [DEFAULT_STREAMING_MODEL]
+	},
+	openAudioStream: async (request) => {
+		const accessToken = await getProviderAuthService().resolveApiKey("xopc-cloud", request.signal);
+		if (!accessToken) throw new Error("XOPC Cloud authorization is unavailable");
+		const source = getModelCatalogStore().getSource("xopc-cloud");
+		const baseUrl = (request.baseUrl ?? source?.baseUrl ?? resolveXopcModelRouterUrl()).replace(/\/+$/, "");
+		const relayUrl = new URL(`${baseUrl}/audio/transcriptions/realtime`);
+		relayUrl.protocol = relayUrl.protocol === "https:" ? "wss:" : "ws:";
+		relayUrl.searchParams.set("model", request.model || DEFAULT_STREAMING_MODEL);
+		return openDashScopeStreamingStt({
+			...request,
+			apiKey: accessToken,
+			baseUrl: relayUrl.toString(),
+			model: request.model || DEFAULT_STREAMING_MODEL
+		});
 	}
 };
 registerMediaUnderstandingProvider(xopcCloudTranscriptionProvider);

package/dist/src/voice/tts/factory.js

@@ -100,6 +100,7 @@ function isTTSAvailable(config) {
 	}
 }
 function isProviderConfigured(provider, config) {
+	if (!getSpeechProvider(provider)) return false;
 	return resolveSpeechProvider(provider, config) !== null;
 }
 function getAvailableProviders(config) {

package/dist/src/voice/tts/openai-compatible-speech.js

@@ -187,7 +187,8 @@ function createOpenAiCompatibleSpeechProvider(options) {
 					response_format: responseFormat,
 					...speed === void 0 ? {} : { speed },
 					...buildExtraJsonBodyFields(config, options.extraJsonBodyFields)
-				}
+				},
+				signal: req.signal
 			});
 			return {
 				audioBuffer: Buffer.from(await response.arrayBuffer()),
@@ -216,6 +217,7 @@ function createOpenAiCompatibleSpeechProvider(options) {
 			const response = await fetchWithTimeoutGuarded(`${baseUrl}/audio/speech`, {
 				timeoutMs: req.timeoutMs,
 				label: options.apiErrorLabel ?? `${options.label} TTS streaming`,
+				signal: req.signal,
 				init: {
 					method: "POST",
 					headers: {

package/dist/src/voice/tts/providers/alibaba-speech.d.ts

@@ -6,11 +6,6 @@
  * doesn't fit the OpenAI-compatible factory.
  *
  * Implementation notes (per docs/voice-rearchitecture.md §8.4.1):
- *   - `synthesizeStream` is intentionally not implemented. Native qwen-tts
- *     streaming uses a WebSocket protocol
- *     (wss://dashscope.aliyuncs.com/api-ws/v1/inference); speak-core's stream
- *     fallback wraps `synthesize` output as a single-chunk ReadableStream so
- *     callers see the same shape.
  *   - `maxTextLength = 512` (DashScope qwen-tts hard limit). Enforced upstream
  *     via `truncateAtSentenceBoundary` before this provider sees the text.
  *   - The audio URL returned by DashScope is hosted on Alibaba's CDN (variable
@@ -24,4 +19,5 @@
  *     as a hardening pass.
  */
 import type { SpeechProviderPlugin } from '../speech-provider-types.js';
+export declare const ALIBABA_REALTIME_TTS_MODEL = "qwen3-tts-flash-realtime";
 export declare const alibabaSpeechProvider: SpeechProviderPlugin;

package/dist/src/voice/tts/providers/alibaba-speech.js

@@ -4,6 +4,7 @@ import { ProviderHttpError } from "../../../media-shared/http/provider-http-erro
 import { fetchWithTimeoutGuarded, postJsonRequest } from "../../../media-shared/http/provider-http.js";
 import "../../../media-shared/http/index.js";
 import { registerSpeechProvider } from "../speech-registry.js";
+import { openDashScopeStreamingTts } from "../../dashscope/streaming-tts-stream.js";
 //#region src/voice/tts/providers/alibaba-speech.ts
 /**
 * Alibaba DashScope TTS provider (qwen-tts model).
@@ -13,11 +14,6 @@ import { registerSpeechProvider } from "../speech-registry.js";
 * doesn't fit the OpenAI-compatible factory.
 *
 * Implementation notes (per docs/voice-rearchitecture.md §8.4.1):
-*   - `synthesizeStream` is intentionally not implemented. Native qwen-tts
-*     streaming uses a WebSocket protocol
-*     (wss://dashscope.aliyuncs.com/api-ws/v1/inference); speak-core's stream
-*     fallback wraps `synthesize` output as a single-chunk ReadableStream so
-*     callers see the same shape.
 *   - `maxTextLength = 512` (DashScope qwen-tts hard limit). Enforced upstream
 *     via `truncateAtSentenceBoundary` before this provider sees the text.
 *   - The audio URL returned by DashScope is hosted on Alibaba's CDN (variable
@@ -34,6 +30,7 @@ init_logger();
 const log = createLogger("SpeechProvider:Alibaba");
 const DEFAULT_BASE_URL = "https://dashscope.aliyuncs.com/api/v1/services/aigc/multimodal-generation/generation";
 const DEFAULT_MODEL = "qwen-tts";
+const ALIBABA_REALTIME_TTS_MODEL = "qwen3-tts-flash-realtime";
 const DEFAULT_VOICE = "longxiaochun";
 const ENV_KEY = "DASHSCOPE_API_KEY";
 const MAX_TEXT_LENGTH = 512;
@@ -135,7 +132,8 @@ const alibabaSpeechProvider = {
 				model,
 				input: { text: req.text },
 				parameters: { voice }
-			}
+			},
+			signal: req.signal
 		})).json();
 		if (data.output?.finish_reason === "null" && !data.output?.audio?.url) throw new Error(`Alibaba TTS API error: ${JSON.stringify(data)}`);
 		let audioBuffer;
@@ -144,7 +142,8 @@ const alibabaSpeechProvider = {
 			const audioResponse = await fetchWithTimeoutGuarded(audioUrl, {
 				timeoutMs: req.timeoutMs,
 				label: "Alibaba TTS audio download",
-				allowPrivateNetwork: false
+				allowPrivateNetwork: false,
+				signal: req.signal
 			});
 			if (!audioResponse.ok) throw new ProviderHttpError({
 				label: "Alibaba TTS audio download",
@@ -166,8 +165,26 @@ const alibabaSpeechProvider = {
 			fileExtension: "wav",
 			voiceCompatible: false
 		};
+	},
+	synthesizeStream: async (req) => {
+		const config = readProviderConfig(req.providerConfig);
+		const apiKey = resolveApiKey(config);
+		if (!apiKey) throw new Error(`Alibaba TTS API key missing (set ${ENV_KEY} or messages.tts.providers.alibaba.apiKey)`);
+		const overrides = req.providerOverrides ?? {};
+		const voice = trimToUndefined(overrides.voice ?? overrides.voiceId) ?? config.voice;
+		const instructions = trimToUndefined(overrides.instructions);
+		return openDashScopeStreamingTts({
+			apiKey,
+			baseUrl: config.baseUrl,
+			model: ALIBABA_REALTIME_TTS_MODEL,
+			voice,
+			text: req.text,
+			...instructions ? { instructions } : {},
+			signal: req.signal,
+			timeoutMs: req.timeoutMs
+		});
 	}
 };
 registerSpeechProvider(alibabaSpeechProvider);
 //#endregion
-export { alibabaSpeechProvider };
+export { ALIBABA_REALTIME_TTS_MODEL, alibabaSpeechProvider };

package/dist/src/voice/tts/providers/edge-speech.js

@@ -94,16 +94,31 @@ function parseDirectiveTokenInternal(ctx) {
 		default: return { handled: false };
 	}
 }
-async function waitForNonEmptyFile(filePath, timeoutMs) {
+async function waitForNonEmptyFile(filePath, timeoutMs, signal) {
 	const deadline = Date.now() + Math.min(Math.max(timeoutMs, 1e3), 5e3);
 	while (Date.now() < deadline) {
+		if (signal.aborted) throw signal.reason;
 		try {
 			if (statSync(filePath).size > 0) return;
 		} catch {}
 		await new Promise((resolve) => setTimeout(resolve, 50));
 	}
 }
-async function synthesizeToBuffer(text, config, voiceOverride, timeoutMs) {
+async function withAbort(promise, signal) {
+	if (signal.aborted) throw signal.reason;
+	return new Promise((resolve, reject) => {
+		const onAbort = () => reject(signal.reason);
+		signal.addEventListener("abort", onAbort, { once: true });
+		promise.then((value) => {
+			signal.removeEventListener("abort", onAbort);
+			resolve(value);
+		}, (error) => {
+			signal.removeEventListener("abort", onAbort);
+			reject(error);
+		});
+	});
+}
+async function synthesizeToBuffer(text, config, voiceOverride, timeoutMs, signal) {
 	const { EdgeTTS } = await import("node-edge-tts");
 	const voice = voiceOverride ?? config.voice;
 	const tempDir = mkdtempSync(path.join(tmpdir(), "tts-edge-"));
@@ -115,7 +130,7 @@ async function synthesizeToBuffer(text, config, voiceOverride, timeoutMs) {
 			outputFormat: config.outputFormat,
 			textLength: text.length
 		}, "Calling Edge TTS");
-		await new EdgeTTS({
+		await withAbort(new EdgeTTS({
 			voice,
 			lang: config.lang,
 			outputFormat: config.outputFormat,
@@ -124,8 +139,8 @@ async function synthesizeToBuffer(text, config, voiceOverride, timeoutMs) {
 			pitch: config.pitch,
 			volume: config.volume,
 			timeout: timeoutMs
-		}).ttsPromise(text, outputPath);
-		await waitForNonEmptyFile(outputPath, timeoutMs);
+		}).ttsPromise(text, outputPath), signal);
+		await waitForNonEmptyFile(outputPath, timeoutMs, signal);
 		const buffer = readFileSync(outputPath);
 		if (buffer.length === 0) throw new Error(`Edge TTS produced an empty ${format} file for voice ${voice}`);
 		return {
@@ -163,29 +178,13 @@ const edgeSpeechProvider = {
 		if (!config.enabled) throw new Error("Edge TTS is disabled in config (messages.tts.providers.edge.enabled = false)");
 		const overrides = req.providerOverrides ?? {};
 		const voiceOverride = trimToUndefined(overrides.voice ?? overrides.voiceId);
-		const { buffer, outputFormat, ext } = await synthesizeToBuffer(req.text, config, voiceOverride, req.timeoutMs);
+		const { buffer, outputFormat, ext } = await synthesizeToBuffer(req.text, config, voiceOverride, req.timeoutMs, req.signal);
 		return {
 			audioBuffer: buffer,
 			outputFormat,
 			fileExtension: ext,
 			voiceCompatible: outputFormat === "opus" || outputFormat === "ogg"
 		};
-	},
-	synthesizeStream: async (req) => {
-		const config = readProviderConfig(req.providerConfig);
-		if (!config.enabled) throw new Error("Edge TTS is disabled in config (messages.tts.providers.edge.enabled = false)");
-		const overrides = req.providerOverrides ?? {};
-		const voiceOverride = trimToUndefined(overrides.voice ?? overrides.voiceId);
-		const { buffer, outputFormat, ext } = await synthesizeToBuffer(req.text, config, voiceOverride, req.timeoutMs);
-		return {
-			audioStream: new ReadableStream({ start(controller) {
-				controller.enqueue(new Uint8Array(buffer));
-				controller.close();
-			} }),
-			outputFormat,
-			fileExtension: ext,
-			voiceCompatible: outputFormat === "opus" || outputFormat === "ogg"
-		};
 	}
 };
 registerSpeechProvider(edgeSpeechProvider);

package/dist/src/voice/tts/providers/minimax-speech.js

@@ -268,19 +268,22 @@ const minimaxSpeechProvider = {
 			model,
 			voiceId: voice,
 			text: req.text,
-			timeoutMs: callTimeoutMs
+			timeoutMs: callTimeoutMs,
+			signal: req.signal
 		});
 		const fileId = await pollTask({
 			baseUrl: config.baseUrl,
 			apiKey,
 			taskId,
-			timeoutMs: callTimeoutMs
+			timeoutMs: callTimeoutMs,
+			signal: req.signal
 		});
 		const audioBuffer = await downloadAudio({
 			baseUrl: config.baseUrl,
 			apiKey,
 			fileId,
-			timeoutMs: callTimeoutMs
+			timeoutMs: callTimeoutMs,
+			signal: req.signal
 		});
 		log.debug({
 			size: audioBuffer.length,

package/dist/src/voice/tts/providers/xopc-cloud-speech.js

@@ -3,6 +3,7 @@ import { init_xopc_cloud_config, resolveXopcModelRouterUrl } from "../../../prov
 import { getProviderAuthService, init_provider_auth_service } from "../../../providers/provider-auth-service.js";
 import { compareCatalogModels } from "../../../providers/model-catalog-ranking.js";
 import { registerSpeechProvider } from "../speech-registry.js";
+import { openDashScopeStreamingTts } from "../../dashscope/streaming-tts-stream.js";
 //#region src/voice/tts/providers/xopc-cloud-speech.ts
 init_model_catalog_store();
 init_provider_auth_service();
@@ -92,7 +93,7 @@ const xopcCloudSpeechProvider = {
 				...speed !== void 0 && catalogModel.tts?.speed ? { speed } : {},
 				...instructions && catalogModel.tts?.instructions ? { instructions } : {}
 			}),
-			signal: AbortSignal.timeout(request.timeoutMs),
+			signal: request.signal,
 			redirect: "error"
 		});
 		if (!response.ok) {
@@ -108,6 +109,30 @@ const xopcCloudSpeechProvider = {
 			fileExtension,
 			voiceCompatible: fileExtension === "opus" || fileExtension === "ogg"
 		};
+	},
+	synthesizeStream: async (request) => {
+		const config = readConfig(request.providerConfig);
+		const model = typeof request.providerOverrides?.model === "string" ? request.providerOverrides.model : config.model;
+		const voice = typeof request.providerOverrides?.voice === "string" ? request.providerOverrides.voice : config.voice;
+		if (!model || !voice) throw new Error("XOPC Cloud realtime TTS model or voice is unavailable");
+		const catalogModel = availableModels().find((entry) => entry.id === model);
+		if (!catalogModel?.tts?.streaming || !catalogModel.tts.outputFormats.includes("pcm")) throw new Error(`XOPC Cloud TTS model does not support PCM streaming: ${model}`);
+		const token = await getProviderAuthService().resolveApiKey("xopc-cloud", request.signal);
+		if (!token) throw new Error("XOPC Cloud authorization is unavailable");
+		const relayUrl = new URL(`${config.baseUrl.replace(/\/+$/, "")}/audio/speech/realtime`);
+		relayUrl.protocol = relayUrl.protocol === "https:" ? "wss:" : "ws:";
+		relayUrl.searchParams.set("model", model);
+		const instructions = typeof request.providerOverrides?.instructions === "string" ? request.providerOverrides.instructions.trim() : config.instructions;
+		return openDashScopeStreamingTts({
+			apiKey: token,
+			baseUrl: relayUrl.toString(),
+			model,
+			voice,
+			text: request.text,
+			...instructions ? { instructions } : {},
+			signal: request.signal,
+			timeoutMs: request.timeoutMs
+		});
 	}
 };
 registerSpeechProvider(xopcCloudSpeechProvider);

package/dist/src/voice/tts/speak-core.d.ts

@@ -23,9 +23,13 @@ export interface SpeakOptions {
     parseDirectives?: boolean;
     modelOverrides?: TTSModelOverrideConfig;
     appConfig?: Config;
-    /** Optional caller-supplied AbortSignal (currently passed through to providers via timeoutMs guard only). */
+    /** Optional caller-supplied cancellation signal. */
     signal?: AbortSignal;
 }
+export interface SpeakStreamOptions extends SpeakOptions {
+    /** Disable provider/model fallback after a realtime route has been selected. */
+    allowFallback?: boolean;
+}
 /**
  * Run the TTS chain (primary → fallback) and return the first successful result.
  */
@@ -51,4 +55,4 @@ export interface SpeakStreamResult {
  * Streaming TTS — tries the chain and returns the first provider's stream.
  * Providers without `synthesizeStream` are wrapped via wrapBufferAsStream.
  */
-export declare function speakStream(text: string, config: TTSConfig, options?: SpeakOptions): Promise<SpeakStreamResult>;
+export declare function speakStream(text: string, config: TTSConfig, options?: SpeakStreamOptions): Promise<SpeakStreamResult>;

package/dist/src/voice/tts/speak-core.js

@@ -87,7 +87,7 @@ async function prepareTextForSynthesis(text, config, options) {
 	};
 }
 /** Build the SpeechSynthesisRequest from a resolved provider + prepared text. */
-function buildSynthesisRequest(resolved, preparedText, ttsOptions, directiveOverrides, appConfig, target) {
+function buildSynthesisRequest(resolved, preparedText, ttsOptions, directiveOverrides, appConfig, target, signal) {
 	const providerOverrides = mergeOverrides(ttsOptions, pickProviderOverrides(directiveOverrides, resolved.providerId));
 	const callTimeoutMs = ttsOptions?.timeoutMs ?? resolved.timeoutMs;
 	return {
@@ -96,6 +96,7 @@ function buildSynthesisRequest(resolved, preparedText, ttsOptions, directiveOver
 		providerConfig: resolved.providerConfig,
 		target,
 		timeoutMs: callTimeoutMs,
+		signal: signal ? AbortSignal.any([signal, AbortSignal.timeout(callTimeoutMs)]) : AbortSignal.timeout(callTimeoutMs),
 		...providerOverrides ? { providerOverrides } : {}
 	};
 }
@@ -145,7 +146,7 @@ async function speak(text, config, options) {
 				textLength: prepared.text.length,
 				provider: resolved.providerId
 			}, "Converting text to speech");
-			const request = buildSynthesisRequest(resolved, prepared.text, prepared.ttsOptions, prepared.directiveOverrides, options?.appConfig, "audio-file");
+			const request = buildSynthesisRequest(resolved, prepared.text, prepared.ttsOptions, prepared.directiveOverrides, options?.appConfig, "audio-file", options?.signal);
 			const result = await resolved.plugin.synthesize(request);
 			const durationSeconds = (Date.now() - startTime) / 1e3;
 			attempts.push({
@@ -209,7 +210,7 @@ async function speakWithProvider(text, config, providerName, options) {
 	if (typeof resolved.plugin.synthesize !== "function") throw new Error(`Provider '${providerName}' does not implement synthesize()`);
 	const prepared = await prepareTextForSynthesis(text, config, options);
 	const startTime = Date.now();
-	const request = buildSynthesisRequest(resolved, prepared.text, prepared.ttsOptions, prepared.directiveOverrides, options?.appConfig, "audio-file");
+	const request = buildSynthesisRequest(resolved, prepared.text, prepared.ttsOptions, prepared.directiveOverrides, options?.appConfig, "audio-file", options?.signal);
 	return {
 		...toTtsResult(resolved, await resolved.plugin.synthesize(request), (Date.now() - startTime) / 1e3),
 		attempts: [{
@@ -229,41 +230,52 @@ async function speakWithProvider(text, config, providerName, options) {
 * Providers without `synthesizeStream` are wrapped via wrapBufferAsStream.
 */
 async function speakStream(text, config, options) {
+	options?.signal?.throwIfAborted();
 	if (!config.enabled) throw new Error("TTS is not enabled");
 	if (config.trigger === "tagged" && !/\[\[tts/.test(text)) throw new Error("TTS trigger is tagged but no [[tts]] directive found");
 	const prepared = await prepareTextForSynthesis(text, config, options);
-	const chain = resolveSpeechProviderChain(config);
+	options?.signal?.throwIfAborted();
+	const chain = options?.allowFallback === false ? [resolveSpeechProvider(prepared.selectedProvider, config)].filter((provider) => provider !== null) : resolveSpeechProviderChain(config);
+	if (chain.length === 0) throw new Error(`TTS provider "${prepared.selectedProvider}" is not available`);
 	let lastError;
-	for (const resolved of chain) try {
-		const baseRequest = buildSynthesisRequest(resolved, prepared.text, prepared.ttsOptions, prepared.directiveOverrides, options?.appConfig, "voice-note");
-		let streamResult;
-		if (typeof resolved.plugin.synthesizeStream === "function") streamResult = await resolved.plugin.synthesizeStream(baseRequest);
-		else if (typeof resolved.plugin.synthesize === "function") streamResult = wrapBufferAsStream(await resolved.plugin.synthesize(baseRequest));
-		else throw new Error(`Provider "${resolved.providerId}" implements neither synthesize nor synthesizeStream`);
-		log.info({
-			provider: resolved.providerId,
-			outputFormat: streamResult.outputFormat,
-			voiceCompatible: streamResult.voiceCompatible,
-			streamed: typeof resolved.plugin.synthesizeStream === "function"
-		}, "TTS stream ready");
-		const release = streamResult.release ?? (async () => {});
-		return {
-			audioStream: streamResult.audioStream,
-			outputFormat: streamResult.outputFormat,
-			fileExtension: streamResult.fileExtension,
-			voiceCompatible: streamResult.voiceCompatible,
-			provider: resolved.providerId,
-			release,
-			ttsText: prepared.text,
-			wasPreprocessed: prepared.wasPreprocessed,
-			wasSummarized: prepared.wasSummarized
-		};
-	} catch (error) {
-		lastError = error instanceof Error ? error : new Error(String(error));
-		log.warn({
-			err: lastError,
-			provider: resolved.providerId
-		}, `TTS stream provider "${resolved.providerId}" failed, trying next`);
+	for (const [index, resolved] of chain.entries()) {
+		options?.signal?.throwIfAborted();
+		try {
+			const baseRequest = buildSynthesisRequest(resolved, prepared.text, prepared.ttsOptions, prepared.directiveOverrides, options?.appConfig, "voice-note", options?.signal);
+			let streamResult;
+			if (typeof resolved.plugin.synthesizeStream === "function") streamResult = await resolved.plugin.synthesizeStream(baseRequest);
+			else if (typeof resolved.plugin.synthesize === "function") streamResult = wrapBufferAsStream(await resolved.plugin.synthesize(baseRequest));
+			else throw new Error(`Provider "${resolved.providerId}" implements neither synthesize nor synthesizeStream`);
+			if (options?.signal?.aborted) {
+				await streamResult.release?.();
+				options.signal.throwIfAborted();
+			}
+			log.info({
+				provider: resolved.providerId,
+				outputFormat: streamResult.outputFormat,
+				voiceCompatible: streamResult.voiceCompatible,
+				streamed: typeof resolved.plugin.synthesizeStream === "function"
+			}, "TTS stream ready");
+			const release = streamResult.release ?? (async () => {});
+			return {
+				audioStream: streamResult.audioStream,
+				outputFormat: streamResult.outputFormat,
+				fileExtension: streamResult.fileExtension,
+				voiceCompatible: streamResult.voiceCompatible,
+				provider: resolved.providerId,
+				release,
+				ttsText: prepared.text,
+				wasPreprocessed: prepared.wasPreprocessed,
+				wasSummarized: prepared.wasSummarized
+			};
+		} catch (error) {
+			options?.signal?.throwIfAborted();
+			lastError = error instanceof Error ? error : new Error(String(error));
+			log.warn({
+				err: lastError,
+				provider: resolved.providerId
+			}, `TTS stream provider "${resolved.providerId}" failed${index < chain.length - 1 ? ", trying next" : ""}`);
+		}
 	}
 	throw lastError ?? /* @__PURE__ */ new Error("All TTS providers failed (streaming)");
 }

package/dist/src/voice/tts/speech-provider-types.d.ts

@@ -86,6 +86,7 @@ export interface SpeechSynthesisRequest {
     providerOverrides?: SpeechProviderOverrides;
     /** Hard timeout from the orchestrator (already merged with provider defaults). */
     timeoutMs: number;
+    signal: AbortSignal;
 }
 export type SpeechSynthesisStreamRequest = SpeechSynthesisRequest;
 /** Voice catalog entry returned by `listVoices`. */

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@xopcai/xopc",
-  "version": "0.0.241",
+  "version": "0.0.243",
   "description": "Local-first personal AI assistant with reviewable user understanding, durable tasks, workflows, automations, and local or cloud models.",
   "type": "module",
   "main": "dist/src/index.js",
@@ -67,10 +67,10 @@
   "dependencies": {
     "@composio/core": "0.14.1",
     "@composio/experimental": "0.2.0",
-    "@earendil-works/pi-agent-core": "0.84.4",
-    "@earendil-works/pi-ai": "0.84.4",
-    "@earendil-works/pi-coding-agent": "0.84.4",
-    "@earendil-works/pi-tui": "0.84.4",
+    "@earendil-works/pi-agent-core": "0.85.1",
+    "@earendil-works/pi-ai": "0.85.1",
+    "@earendil-works/pi-coding-agent": "0.85.1",
+    "@earendil-works/pi-tui": "0.85.1",
     "@grammyjs/runner": "^2.0.3",
     "@hono/node-server": "^2.0.4",
     "@inquirer/prompts": "^8.5.0",
@@ -164,8 +164,8 @@
     "vitest": "4.1.5",
     "vue": "^3.5.35",
     "@xopcai/endpoint-tools-protocol": "0.0.0",
-    "@xopcai/realtime-protocol": "0.0.0",
     "@xopcai/gateway-contract": "0.0.0",
+    "@xopcai/realtime-protocol": "0.0.0",
     "@xopcai/realtime-client": "0.0.0"
   },
   "scripts": {