modelfusion 0.47.3 → 0.49.0

package/model-function/generate-text/streamText.js

@@ -1,133 +1,26 @@
-import { nanoid as createId } from "nanoid";
-import { FunctionEventSource } from "../../core/FunctionEventSource.js";
-import { getGlobalFunctionLogging } from "../../core/GlobalFunctionLogging.js";
-import { getGlobalFunctionObservers } from "../../core/GlobalFunctionObservers.js";
-import { AbortError } from "../../core/api/AbortError.js";
-import { getFunctionCallLogger } from "../../core/getFunctionCallLogger.js";
-import { getRun } from "../../core/getRun.js";
-import { startDurationMeasurement } from "../../util/DurationMeasurement.js";
-import { runSafe } from "../../util/runSafe.js";
 import { AsyncIterableResultPromise } from "../AsyncIterableResultPromise.js";
+import { executeStreamCall } from "../executeStreamCall.js";
 export function streamText(model, prompt, options) {
-    return new AsyncIterableResultPromise(doStreamText(model, prompt, options));
-}
-async function doStreamText(model, prompt, options) {
-    const run = await getRun(options?.run);
-    const eventSource = new FunctionEventSource({
-        observers: [
-            ...getFunctionCallLogger(options?.logging ?? getGlobalFunctionLogging()),
-            ...getGlobalFunctionObservers(),
-            ...(model.settings.observers ?? []),
-            ...(run?.functionObserver != null ? [run.functionObserver] : []),
-            ...(options?.observers ?? []),
-        ],
-        errorHandler: run?.errorHandler,
-    });
-    const durationMeasurement = startDurationMeasurement();
-    const startMetadata = {
+    let accumulatedText = "";
+    let lastFullDelta;
+    return new AsyncIterableResultPromise(executeStreamCall({
         functionType: "text-streaming",
-        callId: `call-${createId()}`,
-        runId: run?.runId,
-        sessionId: run?.sessionId,
-        userId: run?.userId,
-        functionId: options?.functionId,
-        model: model.modelInformation,
-        settings: model.settingsForEvent,
         input: prompt,
-        timestamp: durationMeasurement.startDate,
-        startTimestamp: durationMeasurement.startDate,
-    };
-    eventSource.notify({
-        eventType: "started",
-        ...startMetadata,
-    });
-    const result = await runSafe(async () => {
-        const deltaIterable = await model.doStreamText(prompt, {
-            functionId: options?.functionId,
-            logging: options?.logging,
-            observers: options?.observers,
-            run,
-        });
-        return (async function* () {
-            let accumulatedText = "";
-            let lastFullDelta;
-            for await (const event of deltaIterable) {
-                if (event?.type === "error") {
-                    const error = event.error;
-                    const finishMetadata = {
-                        eventType: "finished",
-                        ...startMetadata,
-                        finishTimestamp: new Date(),
-                        durationInMs: durationMeasurement.durationInMs,
-                    };
-                    eventSource.notify(error instanceof AbortError
-                        ? {
-                            ...finishMetadata,
-                            result: {
-                                status: "abort",
-                            },
-                        }
-                        : {
-                            ...finishMetadata,
-                            result: {
-                                status: "error",
-                                error,
-                            },
-                        });
-                    throw error;
-                }
-                if (event?.type === "delta") {
-                    lastFullDelta = event.fullDelta;
-                    const textDelta = event.valueDelta;
-                    if (textDelta != null && textDelta.length > 0) {
-                        accumulatedText += textDelta;
-                        yield textDelta;
-                    }
-                }
+        model,
+        options,
+        startStream: async (options) => model.doStreamText(prompt, options),
+        processDelta: (delta) => {
+            lastFullDelta = delta.fullDelta;
+            const textDelta = delta.valueDelta;
+            if (textDelta != null && textDelta.length > 0) {
+                accumulatedText += textDelta;
+                return textDelta;
             }
-            const finishMetadata = {
-                eventType: "finished",
-                ...startMetadata,
-                finishTimestamp: new Date(),
-                durationInMs: durationMeasurement.durationInMs,
-            };
-            eventSource.notify({
-                ...finishMetadata,
-                result: {
-                    status: "success",
-                    response: lastFullDelta,
-                    value: accumulatedText,
-                },
-            });
-        })();
-    });
-    if (!result.ok) {
-        const finishMetadata = {
-            eventType: "finished",
-            ...startMetadata,
-            finishTimestamp: new Date(),
-            durationInMs: durationMeasurement.durationInMs,
-        };
-        if (result.isAborted) {
-            eventSource.notify({
-                ...finishMetadata,
-                result: {
-                    status: "abort",
-                },
-            });
-            throw new AbortError();
-        }
-        eventSource.notify({
-            ...finishMetadata,
-            result: {
-                status: "error",
-                error: result.error,
-            },
-        });
-        throw result.error;
-    }
-    return {
-        output: result.value,
-        metadata: startMetadata,
-    };
+            return undefined;
+        },
+        getResult: () => ({
+            response: lastFullDelta,
+            value: accumulatedText,
+        }),
+    }));
 }

package/model-function/generate-text/trimChatPrompt.cjs

@@ -10,7 +10,7 @@ const validateChatPrompt_js_1 = require("./validateChatPrompt.cjs");
  * When the minimal chat prompt (system message + last user message) is already too long, it will only
  * return this minimal chat prompt.
  *
- * @see https://modelfusion.dev/guide/function/generate-text/prompt-format#limiting-the-chat-length
+ * @see https://modelfusion.dev/guide/function/generate-text#limiting-the-chat-length
  */
 async function trimChatPrompt({ prompt, model, tokenLimit = model.contextWindowSize -
     (model.settings.maxCompletionTokens ?? model.contextWindowSize / 4), }) {

package/model-function/generate-text/trimChatPrompt.d.ts

@@ -8,7 +8,7 @@ import { HasContextWindowSize, HasTokenizer, TextGenerationModel, TextGeneration
  * When the minimal chat prompt (system message + last user message) is already too long, it will only
  * return this minimal chat prompt.
  *
- * @see https://modelfusion.dev/guide/function/generate-text/prompt-format#limiting-the-chat-length
+ * @see https://modelfusion.dev/guide/function/generate-text#limiting-the-chat-length
  */
 export declare function trimChatPrompt({ prompt, model, tokenLimit, }: {
     prompt: ChatPrompt;

package/model-function/generate-text/trimChatPrompt.js

@@ -7,7 +7,7 @@ import { validateChatPrompt } from "./validateChatPrompt.js";
  * When the minimal chat prompt (system message + last user message) is already too long, it will only
  * return this minimal chat prompt.
  *
- * @see https://modelfusion.dev/guide/function/generate-text/prompt-format#limiting-the-chat-length
+ * @see https://modelfusion.dev/guide/function/generate-text#limiting-the-chat-length
  */
 export async function trimChatPrompt({ prompt, model, tokenLimit = model.contextWindowSize -
     (model.settings.maxCompletionTokens ?? model.contextWindowSize / 4), }) {

package/model-function/{transcribe-speech/transcribe.cjs → generate-transcription/generateTranscription.cjs}

@@ -1,7 +1,7 @@
 "use strict";
 Object.defineProperty(exports, "__esModule", { value: true });
-exports.transcribe = void 0;
-const executeCall_js_1 = require("../executeCall.cjs");
+exports.generateTranscription = void 0;
+const executeStandardCall_js_1 = require("../executeStandardCall.cjs");
 const ModelFunctionPromise_js_1 = require("../ModelFunctionPromise.cjs");
 /**
  * Transcribe audio data into text.
@@ -9,7 +9,7 @@ const ModelFunctionPromise_js_1 = require("../ModelFunctionPromise.cjs");
  * @example
  * const data = await fs.promises.readFile("data/test.mp3");
  *
- * const transcription = await transcribe(
+ * const transcription = await generateTranscription(
  *   new OpenAITranscriptionModel({ model: "whisper-1" }),
  *   {
  *     type: "mp3",
@@ -17,8 +17,8 @@ const ModelFunctionPromise_js_1 = require("../ModelFunctionPromise.cjs");
  *   }
  * );
  */
-function transcribe(model, data, options) {
-    return new ModelFunctionPromise_js_1.ModelFunctionPromise((0, executeCall_js_1.executeCall)({
+function generateTranscription(model, data, options) {
+    return new ModelFunctionPromise_js_1.ModelFunctionPromise((0, executeStandardCall_js_1.executeStandardCall)({
         functionType: "transcription",
         input: data,
         model,
@@ -32,4 +32,4 @@ function transcribe(model, data, options) {
         },
     }));
 }
-exports.transcribe = transcribe;
+exports.generateTranscription = generateTranscription;

package/model-function/{transcribe-speech/transcribe.d.ts → generate-transcription/generateTranscription.d.ts}

@@ -7,7 +7,7 @@ import { TranscriptionModel, TranscriptionModelSettings } from "./TranscriptionM
  * @example
  * const data = await fs.promises.readFile("data/test.mp3");
  *
- * const transcription = await transcribe(
+ * const transcription = await generateTranscription(
  *   new OpenAITranscriptionModel({ model: "whisper-1" }),
  *   {
  *     type: "mp3",
@@ -15,4 +15,4 @@ import { TranscriptionModel, TranscriptionModelSettings } from "./TranscriptionM
  *   }
  * );
  */
-export declare function transcribe<DATA>(model: TranscriptionModel<DATA, TranscriptionModelSettings>, data: DATA, options?: FunctionOptions): ModelFunctionPromise<string>;
+export declare function generateTranscription<DATA>(model: TranscriptionModel<DATA, TranscriptionModelSettings>, data: DATA, options?: FunctionOptions): ModelFunctionPromise<string>;

package/model-function/{transcribe-speech/transcribe.js → generate-transcription/generateTranscription.js}

@@ -1,4 +1,4 @@
-import { executeCall } from "../executeCall.js";
+import { executeStandardCall } from "../executeStandardCall.js";
 import { ModelFunctionPromise } from "../ModelFunctionPromise.js";
 /**
  * Transcribe audio data into text.
@@ -6,7 +6,7 @@ import { ModelFunctionPromise } from "../ModelFunctionPromise.js";
  * @example
  * const data = await fs.promises.readFile("data/test.mp3");
  *
- * const transcription = await transcribe(
+ * const transcription = await generateTranscription(
  *   new OpenAITranscriptionModel({ model: "whisper-1" }),
  *   {
  *     type: "mp3",
@@ -14,8 +14,8 @@ import { ModelFunctionPromise } from "../ModelFunctionPromise.js";
  *   }
  * );
  */
-export function transcribe(model, data, options) {
-    return new ModelFunctionPromise(executeCall({
+export function generateTranscription(model, data, options) {
+    return new ModelFunctionPromise(executeStandardCall({
         functionType: "transcription",
         input: data,
         model,

package/model-function/index.cjs

@@ -21,9 +21,6 @@ __exportStar(require("./ModelCallMetadata.cjs"), exports);
 __exportStar(require("./ModelInformation.cjs"), exports);
 __exportStar(require("./PromptFormat.cjs"), exports);
 __exportStar(require("./SuccessfulModelCall.cjs"), exports);
-__exportStar(require("./describe-image/ImageDescriptionEvent.cjs"), exports);
-__exportStar(require("./describe-image/ImageDescriptionModel.cjs"), exports);
-__exportStar(require("./describe-image/describeImage.cjs"), exports);
 __exportStar(require("./embed/EmbeddingEvent.cjs"), exports);
 __exportStar(require("./embed/EmbeddingModel.cjs"), exports);
 __exportStar(require("./embed/embed.cjs"), exports);
@@ -31,23 +28,11 @@ __exportStar(require("./generate-image/ImageGenerationEvent.cjs"), exports);
 __exportStar(require("./generate-image/ImageGenerationModel.cjs"), exports);
 __exportStar(require("./generate-image/PromptFormatImageGenerationModel.cjs"), exports);
 __exportStar(require("./generate-image/generateImage.cjs"), exports);
-__exportStar(require("./generate-structure/NoSuchStructureError.cjs"), exports);
-__exportStar(require("./generate-structure/StructureFromTextGenerationModel.cjs"), exports);
-__exportStar(require("./generate-structure/StructureGenerationEvent.cjs"), exports);
-__exportStar(require("./generate-structure/StructureGenerationModel.cjs"), exports);
-__exportStar(require("./generate-structure/StructureOrTextGenerationModel.cjs"), exports);
-__exportStar(require("./generate-structure/StructureParseError.cjs"), exports);
-__exportStar(require("./generate-structure/StructureStreamingEvent.cjs"), exports);
-__exportStar(require("./generate-structure/StructureValidationError.cjs"), exports);
-__exportStar(require("./generate-structure/generateStructure.cjs"), exports);
-__exportStar(require("./generate-structure/generateStructureOrText.cjs"), exports);
-__exportStar(require("./generate-structure/streamStructure.cjs"), exports);
+__exportStar(require("./generate-speech/index.cjs"), exports);
+__exportStar(require("./generate-structure/index.cjs"), exports);
 __exportStar(require("./generate-text/index.cjs"), exports);
-__exportStar(require("./synthesize-speech/SpeechSynthesisEvent.cjs"), exports);
-__exportStar(require("./synthesize-speech/SpeechSynthesisModel.cjs"), exports);
-__exportStar(require("./synthesize-speech/synthesizeSpeech.cjs"), exports);
+__exportStar(require("./generate-transcription/TranscriptionEvent.cjs"), exports);
+__exportStar(require("./generate-transcription/TranscriptionModel.cjs"), exports);
+__exportStar(require("./generate-transcription/generateTranscription.cjs"), exports);
 __exportStar(require("./tokenize-text/Tokenizer.cjs"), exports);
 __exportStar(require("./tokenize-text/countTokens.cjs"), exports);
-__exportStar(require("./transcribe-speech/TranscriptionEvent.cjs"), exports);
-__exportStar(require("./transcribe-speech/TranscriptionModel.cjs"), exports);
-__exportStar(require("./transcribe-speech/transcribe.cjs"), exports);

package/model-function/index.d.ts

@@ -5,9 +5,6 @@ export * from "./ModelCallMetadata.js";
 export * from "./ModelInformation.js";
 export * from "./PromptFormat.js";
 export * from "./SuccessfulModelCall.js";
-export * from "./describe-image/ImageDescriptionEvent.js";
-export * from "./describe-image/ImageDescriptionModel.js";
-export * from "./describe-image/describeImage.js";
 export * from "./embed/EmbeddingEvent.js";
 export * from "./embed/EmbeddingModel.js";
 export * from "./embed/embed.js";
@@ -15,23 +12,11 @@ export * from "./generate-image/ImageGenerationEvent.js";
 export * from "./generate-image/ImageGenerationModel.js";
 export * from "./generate-image/PromptFormatImageGenerationModel.js";
 export * from "./generate-image/generateImage.js";
-export * from "./generate-structure/NoSuchStructureError.js";
-export * from "./generate-structure/StructureFromTextGenerationModel.js";
-export * from "./generate-structure/StructureGenerationEvent.js";
-export * from "./generate-structure/StructureGenerationModel.js";
-export * from "./generate-structure/StructureOrTextGenerationModel.js";
-export * from "./generate-structure/StructureParseError.js";
-export * from "./generate-structure/StructureStreamingEvent.js";
-export * from "./generate-structure/StructureValidationError.js";
-export * from "./generate-structure/generateStructure.js";
-export * from "./generate-structure/generateStructureOrText.js";
-export * from "./generate-structure/streamStructure.js";
+export * from "./generate-speech/index.js";
+export * from "./generate-structure/index.js";
 export * from "./generate-text/index.js";
-export * from "./synthesize-speech/SpeechSynthesisEvent.js";
-export * from "./synthesize-speech/SpeechSynthesisModel.js";
-export * from "./synthesize-speech/synthesizeSpeech.js";
+export * from "./generate-transcription/TranscriptionEvent.js";
+export * from "./generate-transcription/TranscriptionModel.js";
+export * from "./generate-transcription/generateTranscription.js";
 export * from "./tokenize-text/Tokenizer.js";
 export * from "./tokenize-text/countTokens.js";
-export * from "./transcribe-speech/TranscriptionEvent.js";
-export * from "./transcribe-speech/TranscriptionModel.js";
-export * from "./transcribe-speech/transcribe.js";

package/model-function/index.js

@@ -5,9 +5,6 @@ export * from "./ModelCallMetadata.js";
 export * from "./ModelInformation.js";
 export * from "./PromptFormat.js";
 export * from "./SuccessfulModelCall.js";
-export * from "./describe-image/ImageDescriptionEvent.js";
-export * from "./describe-image/ImageDescriptionModel.js";
-export * from "./describe-image/describeImage.js";
 export * from "./embed/EmbeddingEvent.js";
 export * from "./embed/EmbeddingModel.js";
 export * from "./embed/embed.js";
@@ -15,23 +12,11 @@ export * from "./generate-image/ImageGenerationEvent.js";
 export * from "./generate-image/ImageGenerationModel.js";
 export * from "./generate-image/PromptFormatImageGenerationModel.js";
 export * from "./generate-image/generateImage.js";
-export * from "./generate-structure/NoSuchStructureError.js";
-export * from "./generate-structure/StructureFromTextGenerationModel.js";
-export * from "./generate-structure/StructureGenerationEvent.js";
-export * from "./generate-structure/StructureGenerationModel.js";
-export * from "./generate-structure/StructureOrTextGenerationModel.js";
-export * from "./generate-structure/StructureParseError.js";
-export * from "./generate-structure/StructureStreamingEvent.js";
-export * from "./generate-structure/StructureValidationError.js";
-export * from "./generate-structure/generateStructure.js";
-export * from "./generate-structure/generateStructureOrText.js";
-export * from "./generate-structure/streamStructure.js";
+export * from "./generate-speech/index.js";
+export * from "./generate-structure/index.js";
 export * from "./generate-text/index.js";
-export * from "./synthesize-speech/SpeechSynthesisEvent.js";
-export * from "./synthesize-speech/SpeechSynthesisModel.js";
-export * from "./synthesize-speech/synthesizeSpeech.js";
+export * from "./generate-transcription/TranscriptionEvent.js";
+export * from "./generate-transcription/TranscriptionModel.js";
+export * from "./generate-transcription/generateTranscription.js";
 export * from "./tokenize-text/Tokenizer.js";
 export * from "./tokenize-text/countTokens.js";
-export * from "./transcribe-speech/TranscriptionEvent.js";
-export * from "./transcribe-speech/TranscriptionModel.js";
-export * from "./transcribe-speech/transcribe.js";

package/model-provider/elevenlabs/ElevenLabsApiConfiguration.cjs

@@ -18,5 +18,8 @@ class ElevenLabsApiConfiguration extends BaseUrlApiConfiguration_js_1.BaseUrlApi
             throttle,
         });
     }
+    get apiKey() {
+        return this.headers["xi-api-key"];
+    }
 }
 exports.ElevenLabsApiConfiguration = ElevenLabsApiConfiguration;

package/model-provider/elevenlabs/ElevenLabsApiConfiguration.d.ts

@@ -8,4 +8,5 @@ export declare class ElevenLabsApiConfiguration extends BaseUrlApiConfiguration
         retry?: RetryFunction;
         throttle?: ThrottleFunction;
     });
+    get apiKey(): string;
 }

package/model-provider/elevenlabs/ElevenLabsApiConfiguration.js

@@ -15,4 +15,7 @@ export class ElevenLabsApiConfiguration extends BaseUrlApiConfiguration {
             throttle,
         });
     }
+    get apiKey() {
+        return this.headers["xi-api-key"];
+    }
 }

package/model-provider/elevenlabs/ElevenLabsSpeechModel.cjs

@@ -0,0 +1,191 @@
+"use strict";
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.ElevenLabsSpeechModel = void 0;
+const zod_1 = require("zod");
+const callWithRetryAndThrottle_js_1 = require("../../core/api/callWithRetryAndThrottle.cjs");
+const postToApi_js_1 = require("../../core/api/postToApi.cjs");
+const AsyncQueue_js_1 = require("../../event-source/AsyncQueue.cjs");
+const AbstractModel_js_1 = require("../../model-function/AbstractModel.cjs");
+const SimpleWebSocket_js_1 = require("../../util/SimpleWebSocket.cjs");
+const parseJSON_js_1 = require("../../util/parseJSON.cjs");
+const ElevenLabsApiConfiguration_js_1 = require("./ElevenLabsApiConfiguration.cjs");
+const ElevenLabsError_js_1 = require("./ElevenLabsError.cjs");
+const elevenLabsModels = [
+    "eleven_multilingual_v2",
+    "eleven_multilingual_v1",
+    "eleven_monolingual_v1",
+];
+const defaultModel = "eleven_multilingual_v2";
+/**
+ * Synthesize speech using the ElevenLabs Text to Speech API.
+ *
+ * @see https://api.elevenlabs.io/docs#/text-to-speech/Text_to_speech_v1_text_to_speech__voice_id__post
+ */
+class ElevenLabsSpeechModel extends AbstractModel_js_1.AbstractModel {
+    constructor(settings) {
+        super({ settings });
+        Object.defineProperty(this, "provider", {
+            enumerable: true,
+            configurable: true,
+            writable: true,
+            value: "elevenlabs"
+        });
+    }
+    get modelName() {
+        return this.settings.voice;
+    }
+    async callAPI(text, options) {
+        return (0, callWithRetryAndThrottle_js_1.callWithRetryAndThrottle)({
+            retry: this.settings.api?.retry,
+            throttle: this.settings.api?.throttle,
+            call: async () => callElevenLabsTextToSpeechAPI({
+                api: this.settings.api,
+                abortSignal: options?.run?.abortSignal,
+                text,
+                voiceId: this.settings.voice,
+                modelId: this.settings.model,
+                voiceSettings: this.settings.voiceSettings,
+            }),
+        });
+    }
+    get settingsForEvent() {
+        return {
+            model: this.settings.model,
+            voice: this.settings.voice,
+            voiceSettings: this.settings.voiceSettings,
+        };
+    }
+    doGenerateSpeechStandard(text, options) {
+        return this.callAPI(text, options);
+    }
+    async doGenerateSpeechStreamDuplex(textStream
+    // options?: FunctionOptions | undefined
+    ) {
+        const responseSchema = zod_1.z.union([
+            zod_1.z.object({
+                audio: zod_1.z.string(),
+                isFinal: zod_1.z.literal(false).nullable(),
+                normalizedAlignment: zod_1.z
+                    .object({
+                    chars: zod_1.z.array(zod_1.z.string()),
+                    charStartTimesMs: zod_1.z.array(zod_1.z.number()),
+                    charDurationsMs: zod_1.z.array(zod_1.z.number()),
+                })
+                    .nullable(),
+            }),
+            zod_1.z.object({
+                isFinal: zod_1.z.literal(true),
+            }),
+            zod_1.z.object({
+                message: zod_1.z.string(),
+                error: zod_1.z.string(),
+                code: zod_1.z.number(),
+            }),
+        ]);
+        const queue = new AsyncQueue_js_1.AsyncQueue();
+        const model = this.settings.model ?? defaultModel;
+        const socket = await (0, SimpleWebSocket_js_1.createSimpleWebSocket)(`wss://api.elevenlabs.io/v1/text-to-speech/${this.settings.voice}/stream-input?model_id=${model}`);
+        socket.onopen = async () => {
+            const api = this.settings.api ?? new ElevenLabsApiConfiguration_js_1.ElevenLabsApiConfiguration();
+            // send begin-of-stream (BOS) message:
+            socket.send(JSON.stringify({
+                // The JS WebSocket API does not support authorization headers, so we send the API key in the BOS message.
+                // See https://stackoverflow.com/questions/4361173/http-headers-in-websockets-client-api
+                xi_api_key: api.apiKey,
+                text: " ",
+                voice_settings: toApiVoiceSettings(this.settings.voiceSettings),
+                generation_config: toGenerationConfig(this.settings.generationConfig),
+            }));
+            // send text in chunks:
+            let textBuffer = "";
+            for await (const textDelta of textStream) {
+                textBuffer += textDelta;
+                // using ". " as separator: sending in full sentences improves the quality
+                // of the audio output significantly.
+                const separator = textBuffer.lastIndexOf(". ");
+                if (separator === -1) {
+                    continue;
+                }
+                const textToProcess = textBuffer.slice(0, separator);
+                textBuffer = textBuffer.slice(separator + 1);
+                socket.send(JSON.stringify({
+                    text: textToProcess,
+                    try_trigger_generation: true,
+                }));
+            }
+            // send remaining text:
+            if (textBuffer.length > 0) {
+                socket.send(JSON.stringify({
+                    text: `${textBuffer} `,
+                    try_trigger_generation: true,
+                }));
+            }
+            // send end-of-stream (EOS) message:
+            socket.send(JSON.stringify({ text: "" }));
+        };
+        socket.onmessage = (event) => {
+            const parseResult = (0, parseJSON_js_1.safeParseJsonWithZod)(event.data, responseSchema);
+            if (!parseResult.success) {
+                queue.push({ type: "error", error: parseResult.error });
+                return;
+            }
+            const response = parseResult.data;
+            if ("error" in response) {
+                queue.push({ type: "error", error: response });
+                return;
+            }
+            if (!response.isFinal) {
+                queue.push({
+                    type: "delta",
+                    fullDelta: event,
+                    valueDelta: Buffer.from(response.audio, "base64"),
+                });
+            }
+        };
+        socket.onerror = (error) => {
+            queue.push({ type: "error", error });
+        };
+        socket.onclose = () => {
+            queue.close();
+        };
+        return queue;
+    }
+    withSettings(additionalSettings) {
+        return new ElevenLabsSpeechModel({
+            ...this.settings,
+            ...additionalSettings,
+        });
+    }
+}
+exports.ElevenLabsSpeechModel = ElevenLabsSpeechModel;
+async function callElevenLabsTextToSpeechAPI({ api = new ElevenLabsApiConfiguration_js_1.ElevenLabsApiConfiguration(), abortSignal, text, voiceId, modelId, voiceSettings, }) {
+    return (0, postToApi_js_1.postJsonToApi)({
+        url: api.assembleUrl(`/text-to-speech/${voiceId}`),
+        headers: api.headers,
+        body: {
+            text,
+            model_id: modelId ?? defaultModel,
+            voice_settings: toApiVoiceSettings(voiceSettings),
+        },
+        failedResponseHandler: ElevenLabsError_js_1.failedElevenLabsCallResponseHandler,
+        successfulResponseHandler: (0, postToApi_js_1.createAudioMpegResponseHandler)(),
+        abortSignal,
+    });
+}
+function toApiVoiceSettings(voiceSettings) {
+    return voiceSettings != null
+        ? {
+            stability: voiceSettings.stability,
+            similarity_boost: voiceSettings.similarityBoost,
+            style: voiceSettings.style,
+            use_speaker_boost: voiceSettings.useSpeakerBoost,
+        }
+        : undefined;
+}
+function toGenerationConfig(generationConfig) {
+    return generationConfig != null
+        ? {
+            chunk_length_schedule: generationConfig.chunkLengthSchedule,
+        }
+        : undefined;
+}

package/model-provider/elevenlabs/ElevenLabsSpeechModel.d.ts

@@ -0,0 +1,39 @@
+/// <reference types="node" />
+import { FunctionOptions } from "../../core/FunctionOptions.js";
+import { ApiConfiguration } from "../../core/api/ApiConfiguration.js";
+import { AbstractModel } from "../../model-function/AbstractModel.js";
+import { Delta } from "../../model-function/Delta.js";
+import { StreamingSpeechGenerationModel, SpeechGenerationModelSettings } from "../../model-function/generate-speech/SpeechGenerationModel.js";
+declare const elevenLabsModels: readonly ["eleven_multilingual_v2", "eleven_multilingual_v1", "eleven_monolingual_v1"];
+export interface ElevenLabsSpeechModelSettings extends SpeechGenerationModelSettings {
+    api?: ApiConfiguration & {
+        apiKey: string;
+    };
+    voice: string;
+    model?: (typeof elevenLabsModels)[number] | (string & {});
+    voiceSettings?: {
+        stability: number;
+        similarityBoost: number;
+        style?: number;
+        useSpeakerBoost?: boolean;
+    };
+    generationConfig?: {
+        chunkLengthSchedule: number[];
+    };
+}
+/**
+ * Synthesize speech using the ElevenLabs Text to Speech API.
+ *
+ * @see https://api.elevenlabs.io/docs#/text-to-speech/Text_to_speech_v1_text_to_speech__voice_id__post
+ */
+export declare class ElevenLabsSpeechModel extends AbstractModel<ElevenLabsSpeechModelSettings> implements StreamingSpeechGenerationModel<ElevenLabsSpeechModelSettings> {
+    constructor(settings: ElevenLabsSpeechModelSettings);
+    readonly provider = "elevenlabs";
+    get modelName(): string;
+    private callAPI;
+    get settingsForEvent(): Partial<ElevenLabsSpeechModelSettings>;
+    doGenerateSpeechStandard(text: string, options?: FunctionOptions): Promise<Buffer>;
+    doGenerateSpeechStreamDuplex(textStream: AsyncIterable<string>): Promise<AsyncIterable<Delta<Buffer>>>;
+    withSettings(additionalSettings: Partial<ElevenLabsSpeechModelSettings>): this;
+}
+export {};