From 0cd041dc1a2866e714c9200bbb4ced88d79e145f Mon Sep 17 00:00:00 2001
From: luke-speechify <289678208+luke-speechify@users.noreply.github.com>
Date: Mon, 6 Jul 2026 13:17:10 +0100
Subject: [PATCH 01/13] feat(speechify): add Speechify TTS plugin

Add @livekit/agents-plugin-speechify: streaming text-to-speech with
word-level timestamps via the official @speechify/api SDK. Chunks
streamed input into sequential /audio/speech calls, emitting audio and
timed transcripts per sentence (streaming + alignedTranscript). Maintained
by Speechify (Speechify-AI/livekit-plugin-js).
---
 README.md                            |   1 +
 plugins/speechify/CHANGELOG.md       |   1 +
 plugins/speechify/README.md          |  21 ++
 plugins/speechify/api-extractor.json |  20 ++
 plugins/speechify/package.json       |  52 +++++
 plugins/speechify/src/index.ts       |  19 ++
 plugins/speechify/src/models.ts      |   5 +
 plugins/speechify/src/tts.ts         | 279 +++++++++++++++++++++++++++
 plugins/speechify/tsconfig.json      |  16 ++
 plugins/speechify/tsup.config.ts     |   7 +
 pnpm-lock.yaml                       |  48 ++++-
 turbo.json                           |   1 +
 12 files changed, 463 insertions(+), 7 deletions(-)
 create mode 100644 plugins/speechify/CHANGELOG.md
 create mode 100644 plugins/speechify/README.md
 create mode 100644 plugins/speechify/api-extractor.json
 create mode 100644 plugins/speechify/package.json
 create mode 100644 plugins/speechify/src/index.ts
 create mode 100644 plugins/speechify/src/models.ts
 create mode 100644 plugins/speechify/src/tts.ts
 create mode 100644 plugins/speechify/tsconfig.json
 create mode 100644 plugins/speechify/tsup.config.ts

diff --git a/README.md b/README.md
index e708dcf6e..d3aa9b27f 100644
--- a/README.md
+++ b/README.md
@@ -70,6 +70,7 @@ Currently, only the following plugins are supported:
 | [@livekit/agents-plugin-resemble](https://www.npmjs.com/package/@livekit/agents-plugin-resemble)     | TTS           |
 | [@livekit/agents-plugin-rime](https://www.npmjs.com/package/@livekit/agents-plugin-rime)             | TTS           |
 | [@livekit/agents-plugin-inworld](https://www.npmjs.com/package/@livekit/agents-plugin-inworld)       | STT, TTS      |
+| [@livekit/agents-plugin-speechify](https://www.npmjs.com/package/@livekit/agents-plugin-speechify)   | TTS           |
 | [@livekit/agents-plugin-silero](https://www.npmjs.com/package/@livekit/agents-plugin-silero)         | VAD           |
 | [@livekit/agents-plugin-livekit](https://www.npmjs.com/package/@livekit/agents-plugin-livekit)       | EOU           |
 | [@livekit/agents-plugin-anam](https://www.npmjs.com/package/@livekit/agents-plugin-anam)             | Avatar        |
diff --git a/plugins/speechify/CHANGELOG.md b/plugins/speechify/CHANGELOG.md
new file mode 100644
index 000000000..62cf44533
--- /dev/null
+++ b/plugins/speechify/CHANGELOG.md
@@ -0,0 +1 @@
+# @livekit/agents-plugin-speechify
diff --git a/plugins/speechify/README.md b/plugins/speechify/README.md
new file mode 100644
index 000000000..4fe5f6dcf
--- /dev/null
+++ b/plugins/speechify/README.md
@@ -0,0 +1,21 @@
+<!--
+SPDX-FileCopyrightText: 2025 Speechify, Inc.
+
+SPDX-License-Identifier: Apache-2.0
+-->
+# Speechify plugin for LiveKit Agents
+
+The Agents Framework is designed for building realtime, programmable
+participants that run on servers. Use it to create conversational, multi-modal
+voice agents that can see, hear, and understand.
+
+This package contains the Speechify plugin, which provides streaming text-to-speech
+with word-level timestamps. Refer to the [documentation](https://docs.livekit.io/agents/overview/)
+for information on how to use it, or browse the [API
+reference](https://docs.livekit.io/agents-js/modules/plugins_agents_plugin_speechify.html).
+See the [repository](https://github.com/livekit/agents-js) for more information
+about the framework as a whole.
+
+This plugin is maintained by Speechify. Issues and contributions specific to the
+plugin are welcome at
+[Speechify-AI/livekit-plugin-js](https://github.com/Speechify-AI/livekit-plugin-js).
diff --git a/plugins/speechify/api-extractor.json b/plugins/speechify/api-extractor.json
new file mode 100644
index 000000000..1f75e0708
--- /dev/null
+++ b/plugins/speechify/api-extractor.json
@@ -0,0 +1,20 @@
+/**
+ * Config file for API Extractor.  For more info, please visit: https://api-extractor.com
+ */
+{
+  "$schema": "https://developer.microsoft.com/json-schemas/api-extractor/v7/api-extractor.schema.json",
+
+  /**
+   * Optionally specifies another JSON config file that this file extends from.  This provides a way for
+   * standard settings to be shared across multiple projects.
+   *
+   * If the path starts with "./" or "../", the path is resolved relative to the folder of the file that contains
+   * the "extends" field.  Otherwise, the first path segment is interpreted as an NPM package name, and will be
+   * resolved using NodeJS require().
+   *
+   * SUPPORTED TOKENS: none
+   * DEFAULT VALUE: ""
+   */
+  "extends": "../../api-extractor-shared.json",
+  "mainEntryPointFilePath": "./dist/index.d.ts"
+}
diff --git a/plugins/speechify/package.json b/plugins/speechify/package.json
new file mode 100644
index 000000000..fbf45fcac
--- /dev/null
+++ b/plugins/speechify/package.json
@@ -0,0 +1,52 @@
+{
+  "name": "@livekit/agents-plugin-speechify",
+  "version": "1.5.0",
+  "description": "Speechify plugin for LiveKit Node Agents",
+  "main": "dist/index.js",
+  "require": "dist/index.cjs",
+  "types": "dist/index.d.ts",
+  "exports": {
+    "import": {
+      "types": "./dist/index.d.ts",
+      "default": "./dist/index.js"
+    },
+    "require": {
+      "types": "./dist/index.d.cts",
+      "default": "./dist/index.cjs"
+    }
+  },
+  "author": "Speechify",
+  "type": "module",
+  "repository": "git@github.com:livekit/agents-js.git",
+  "license": "Apache-2.0",
+  "files": [
+    "dist",
+    "src",
+    "README.md"
+  ],
+  "scripts": {
+    "build": "tsup --onSuccess \"pnpm build:types\"",
+    "build:types": "tsc --declaration --emitDeclarationOnly && node ../../scripts/copyDeclarationOutput.js",
+    "clean": "rm -rf dist",
+    "clean:build": "pnpm clean && pnpm build",
+    "lint": "eslint -f unix \"src/**/*.{ts,js}\"",
+    "api:check": "api-extractor run --typescript-compiler-folder ../../node_modules/typescript",
+    "api:update": "api-extractor run --local --typescript-compiler-folder ../../node_modules/typescript --verbose"
+  },
+  "devDependencies": {
+    "@livekit/agents": "workspace:*",
+    "@livekit/agents-plugin-openai": "workspace:*",
+    "@livekit/agents-plugins-test": "workspace:*",
+    "@livekit/rtc-node": "catalog:",
+    "@microsoft/api-extractor": "^7.35.0",
+    "tsup": "^8.3.5",
+    "typescript": "^5.0.0"
+  },
+  "dependencies": {
+    "@speechify/api": "^2.0.0"
+  },
+  "peerDependencies": {
+    "@livekit/agents": "workspace:*",
+    "@livekit/rtc-node": "catalog:"
+  }
+}
diff --git a/plugins/speechify/src/index.ts b/plugins/speechify/src/index.ts
new file mode 100644
index 000000000..252e901dd
--- /dev/null
+++ b/plugins/speechify/src/index.ts
@@ -0,0 +1,19 @@
+// SPDX-FileCopyrightText: 2025 LiveKit, Inc.
+//
+// SPDX-License-Identifier: Apache-2.0
+import { Plugin } from '@livekit/agents';
+
+export * from './models.js';
+export * from './tts.js';
+
+class SpeechifyPlugin extends Plugin {
+  constructor() {
+    super({
+      title: 'speechify',
+      version: __PACKAGE_VERSION__,
+      package: __PACKAGE_NAME__,
+    });
+  }
+}
+
+Plugin.registerPlugin(new SpeechifyPlugin());
diff --git a/plugins/speechify/src/models.ts b/plugins/speechify/src/models.ts
new file mode 100644
index 000000000..8c4b54a69
--- /dev/null
+++ b/plugins/speechify/src/models.ts
@@ -0,0 +1,5 @@
+// SPDX-FileCopyrightText: 2025 Speechify, Inc.
+//
+// SPDX-License-Identifier: Apache-2.0
+
+export type TTSModels = 'simba-english' | 'simba-multilingual' | 'simba-3.0';
diff --git a/plugins/speechify/src/tts.ts b/plugins/speechify/src/tts.ts
new file mode 100644
index 000000000..032119e67
--- /dev/null
+++ b/plugins/speechify/src/tts.ts
@@ -0,0 +1,279 @@
+// SPDX-FileCopyrightText: 2025 Speechify, Inc.
+//
+// SPDX-License-Identifier: Apache-2.0
+import {
+  type APIConnectOptions,
+  APIConnectionError,
+  APIStatusError,
+  AudioByteStream,
+  USERDATA_TIMED_TRANSCRIPT,
+  createTimedString,
+  shortuuid,
+  tokenize,
+  tts,
+} from '@livekit/agents';
+import type { Speechify } from '@speechify/api';
+import { SpeechifyClient, SpeechifyError } from '@speechify/api';
+import type { TTSModels } from './models.js';
+
+const NUM_CHANNELS = 1;
+const SAMPLE_RATE = 24000;
+const AUDIO_FORMAT: Speechify.GetSpeechRequest.AudioFormat = 'pcm';
+const DEFAULT_VOICE_ID = 'jack';
+
+export interface TTSOptions {
+  voiceId: string;
+  model?: TTSModels;
+  language?: string;
+  loudnessNormalization?: boolean;
+  textNormalization?: boolean;
+  apiKey?: string;
+  baseUrl?: string;
+  client?: SpeechifyClient;
+  tokenizer?: tokenize.SentenceTokenizer;
+}
+
+const defaultOptions = (): Omit<TTSOptions, 'client' | 'tokenizer'> => ({
+  voiceId: DEFAULT_VOICE_ID,
+});
+
+const buildSpeechRequest = (text: string, opts: TTSOptions): Speechify.GetSpeechRequest => {
+  const request: Speechify.GetSpeechRequest = {
+    audio_format: AUDIO_FORMAT,
+    input: text,
+    voice_id: opts.voiceId,
+  };
+  if (opts.model) request.model = opts.model;
+  if (opts.language) request.language = opts.language;
+  if (opts.loudnessNormalization !== undefined || opts.textNormalization !== undefined) {
+    request.options = {
+      loudness_normalization: opts.loudnessNormalization,
+      text_normalization: opts.textNormalization,
+    };
+  }
+  return request;
+};
+
+const toError = (e: unknown): Error => {
+  if (e instanceof SpeechifyError) {
+    return new APIStatusError({ message: e.message, options: { statusCode: e.statusCode ?? -1 } });
+  }
+  return new APIConnectionError({ message: e instanceof Error ? e.message : String(e) });
+};
+
+export class TTS extends tts.TTS {
+  label = 'speechify.TTS';
+  #opts: TTSOptions;
+  #client: SpeechifyClient;
+  #tokenizer: tokenize.SentenceTokenizer;
+
+  /**
+   * Create a new instance of Speechify TTS.
+   *
+   * @remarks
+   * `apiKey` must be set, either via the constructor or the `SPEECHIFY_API_KEY`
+   * environment variable. Pass a preconfigured `client` to reuse an existing
+   * `SpeechifyClient` (in which case `apiKey`/`baseUrl` are ignored).
+   *
+   * Synthesis uses the Speechify `/audio/speech` endpoint, which returns raw PCM
+   * (24 kHz mono) plus word-level speech marks. `stream()` chunks input into
+   * sentences and issues one request per sentence, emitting audio and aligned
+   * word timestamps as each sentence completes.
+   */
+  constructor(opts: Partial<TTSOptions> = {}) {
+    const merged = { ...defaultOptions(), ...opts };
+
+    super(SAMPLE_RATE, NUM_CHANNELS, { streaming: true, alignedTranscript: true });
+
+    this.#opts = merged;
+    this.#tokenizer = merged.tokenizer ?? new tokenize.basic.SentenceTokenizer();
+
+    if (merged.client) {
+      this.#client = merged.client;
+    } else {
+      const apiKey = merged.apiKey ?? process.env.SPEECHIFY_API_KEY;
+      if (!apiKey) {
+        throw new Error(
+          'Speechify API key is required, whether as an argument or as $SPEECHIFY_API_KEY',
+        );
+      }
+      this.#client = new SpeechifyClient({ apiKey, baseUrl: merged.baseUrl });
+    }
+  }
+
+  get model(): string {
+    return this.#opts.model ?? 'unknown';
+  }
+
+  get provider(): string {
+    return 'Speechify';
+  }
+
+  get client(): SpeechifyClient {
+    return this.#client;
+  }
+
+  get options(): TTSOptions {
+    return this.#opts;
+  }
+
+  get tokenizer(): tokenize.SentenceTokenizer {
+    return this.#tokenizer;
+  }
+
+  updateOptions(opts: Partial<Omit<TTSOptions, 'client' | 'apiKey' | 'baseUrl' | 'tokenizer'>>) {
+    this.#opts = { ...this.#opts, ...opts };
+  }
+
+  synthesize(text: string, connOptions?: APIConnectOptions): tts.ChunkedStream {
+    return new ChunkedStream(this, text, this.#opts, connOptions);
+  }
+
+  stream(options?: { connOptions?: APIConnectOptions }): tts.SynthesizeStream {
+    return new SynthesizeStream(this, this.#opts, options?.connOptions);
+  }
+}
+
+const timedStringsFromMarks = (
+  marks: Speechify.SpeechMarks | undefined,
+  offsetSeconds: number,
+): ReturnType<typeof createTimedString>[] => {
+  if (!marks?.chunks) return [];
+  const out: ReturnType<typeof createTimedString>[] = [];
+  for (const chunk of marks.chunks) {
+    if (!chunk.value || chunk.start_time === undefined) continue;
+    out.push(
+      createTimedString({
+        text: chunk.value,
+        startTime: chunk.start_time / 1000 + offsetSeconds,
+        endTime: chunk.end_time !== undefined ? chunk.end_time / 1000 + offsetSeconds : undefined,
+      }),
+    );
+  }
+  return out;
+};
+
+export class ChunkedStream extends tts.ChunkedStream {
+  label = 'speechify.ChunkedStream';
+  #opts: TTSOptions;
+  #tts: TTS;
+  #timeoutInSeconds?: number;
+
+  constructor(ttsInstance: TTS, text: string, opts: TTSOptions, connOptions?: APIConnectOptions) {
+    super(text, ttsInstance, connOptions);
+    this.#tts = ttsInstance;
+    this.#opts = opts;
+    this.#timeoutInSeconds =
+      connOptions?.timeoutMs !== undefined ? connOptions.timeoutMs / 1000 : undefined;
+  }
+
+  protected async run() {
+    const requestId = shortuuid();
+    const bstream = new AudioByteStream(SAMPLE_RATE, NUM_CHANNELS);
+
+    try {
+      const response = await this.#tts.client.audio.speech(
+        buildSpeechRequest(this.inputText, this.#opts),
+        { abortSignal: this.abortSignal, timeoutInSeconds: this.#timeoutInSeconds },
+      );
+
+      const audio = Buffer.from(response.audio_data, 'base64');
+      const timed = timedStringsFromMarks(response.speech_marks, 0);
+      let attached = false;
+
+      const putFrames = (frames: ReturnType<AudioByteStream['write']>) => {
+        for (const frame of frames) {
+          if (!attached && timed.length > 0) {
+            frame.userdata[USERDATA_TIMED_TRANSCRIPT] = timed;
+            attached = true;
+          }
+          this.queue.put({ requestId, frame, final: false, segmentId: requestId });
+        }
+      };
+
+      putFrames(bstream.write(audio));
+      putFrames(bstream.flush());
+      this.queue.close();
+    } catch (e) {
+      if (!this.queue.closed) this.queue.close();
+      if (this.abortSignal.aborted) return;
+      throw toError(e);
+    }
+  }
+}
+
+export class SynthesizeStream extends tts.SynthesizeStream {
+  label = 'speechify.SynthesizeStream';
+  #opts: TTSOptions;
+  #tts: TTS;
+
+  constructor(ttsInstance: TTS, opts: TTSOptions, connOptions?: APIConnectOptions) {
+    super(ttsInstance, connOptions);
+    this.#tts = ttsInstance;
+    this.#opts = opts;
+  }
+
+  protected async run() {
+    const sentenceStream = this.#tts.tokenizer.stream();
+    let cumulativeDuration = 0;
+
+    const forwardInput = async () => {
+      for await (const input of this.input) {
+        if (this.abortController.signal.aborted) break;
+        if (input === SynthesizeStream.FLUSH_SENTINEL) {
+          sentenceStream.flush();
+        } else {
+          sentenceStream.pushText(input);
+        }
+      }
+      sentenceStream.endInput();
+      sentenceStream.close();
+    };
+
+    const synthesizeSentence = async (text: string) => {
+      const requestId = shortuuid();
+      const bstream = new AudioByteStream(SAMPLE_RATE, NUM_CHANNELS);
+
+      const response = await this.#tts.client.audio.speech(buildSpeechRequest(text, this.#opts), {
+        abortSignal: this.abortSignal,
+        timeoutInSeconds: this.connOptions.timeoutMs / 1000,
+      });
+
+      const audio = Buffer.from(response.audio_data, 'base64');
+      const timed = timedStringsFromMarks(response.speech_marks, cumulativeDuration);
+      let attached = false;
+
+      const putFrames = (frames: ReturnType<AudioByteStream['write']>) => {
+        for (const frame of frames) {
+          if (!attached && timed.length > 0) {
+            frame.userdata[USERDATA_TIMED_TRANSCRIPT] = timed;
+            attached = true;
+          }
+          cumulativeDuration += frame.samplesPerChannel / frame.sampleRate;
+          this.queue.put({ requestId, frame, final: false, segmentId: requestId });
+        }
+      };
+
+      putFrames(bstream.write(audio));
+      putFrames(bstream.flush());
+    };
+
+    const consume = async () => {
+      for await (const ev of sentenceStream) {
+        if (this.abortController.signal.aborted) break;
+        const text = ev.token.trim();
+        if (!text) continue;
+        await synthesizeSentence(text);
+      }
+      this.queue.put(SynthesizeStream.END_OF_STREAM);
+    };
+
+    try {
+      await Promise.all([forwardInput(), consume()]);
+    } catch (e) {
+      if (!this.queue.closed) this.queue.close();
+      if (this.abortSignal.aborted) return;
+      throw toError(e);
+    }
+  }
+}
diff --git a/plugins/speechify/tsconfig.json b/plugins/speechify/tsconfig.json
new file mode 100644
index 000000000..eed8cb31e
--- /dev/null
+++ b/plugins/speechify/tsconfig.json
@@ -0,0 +1,16 @@
+{
+  "extends": "../../tsconfig.json",
+  "include": ["./src"],
+  "compilerOptions": {
+    // match output dir to input dir. e.g. dist/index instead of dist/src/index
+    "rootDir": "./src",
+    "declarationDir": "./dist",
+    "outDir": "./dist"
+  },
+  "typedocOptions": {
+    "name": "plugins/agents-plugin-speechify",
+    "entryPointStrategy": "resolve",
+    "readme": "none",
+    "entryPoints": ["src/index.ts"]
+  }
+}
diff --git a/plugins/speechify/tsup.config.ts b/plugins/speechify/tsup.config.ts
new file mode 100644
index 000000000..8ca20961f
--- /dev/null
+++ b/plugins/speechify/tsup.config.ts
@@ -0,0 +1,7 @@
+import { defineConfig } from 'tsup';
+
+import defaults from '../../tsup.config';
+
+export default defineConfig({
+  ...defaults,
+});
diff --git a/pnpm-lock.yaml b/pnpm-lock.yaml
index 3f01e7345..10262223b 100644
--- a/pnpm-lock.yaml
+++ b/pnpm-lock.yaml
@@ -1327,6 +1327,34 @@ importers:
         specifier: ^5.0.0
         version: 5.9.3
 
+  plugins/speechify:
+    dependencies:
+      '@speechify/api':
+        specifier: ^2.0.0
+        version: 2.0.0
+    devDependencies:
+      '@livekit/agents':
+        specifier: workspace:*
+        version: link:../../agents
+      '@livekit/agents-plugin-openai':
+        specifier: workspace:*
+        version: link:../openai
+      '@livekit/agents-plugins-test':
+        specifier: workspace:*
+        version: link:../test
+      '@livekit/rtc-node':
+        specifier: 'catalog:'
+        version: 0.13.30
+      '@microsoft/api-extractor':
+        specifier: ^7.35.0
+        version: 7.43.7(@types/node@25.6.0)
+      tsup:
+        specifier: ^8.3.5
+        version: 8.4.0(@microsoft/api-extractor@7.43.7(@types/node@25.6.0))(postcss@8.5.9)(tsx@4.21.0)(typescript@5.9.3)
+      typescript:
+        specifier: ^5.0.0
+        version: 5.9.3
+
   plugins/tavus:
     dependencies:
       livekit-server-sdk:
@@ -2713,6 +2741,10 @@ packages:
   '@rushstack/ts-command-line@4.21.0':
     resolution: {integrity: sha512-z38FLUCn8M9FQf19gJ9eltdwkvc47PxvJmVZS6aKwbBAa3Pis3r3A+ZcBCVPNb9h/Tbga+i0tHdzoSGUoji9GQ==}
 
+  '@speechify/api@2.0.0':
+    resolution: {integrity: sha512-GYIYO/B0bRArg9e4ljLxHL/ZG27p7X68UGu6bvEqBk0I8CSpqPKHO+rsGGJR+oIRzXZpR5LVNcS6RPN0ljnvNw==}
+    engines: {node: '>=18.0.0'}
+
   '@standard-schema/spec@1.1.0':
     resolution: {integrity: sha512-l2aFy5jALhniG5HgqrD6jXLi/rUWrKvqN/qJx6yoJsgKhblVd+iqqU4RCXavm/jPityDo5TCvKMnpjKnOriy0w==}
 
@@ -5678,7 +5710,7 @@ snapshots:
   '@eslint/eslintrc@2.1.4':
     dependencies:
       ajv: 6.12.6
-      debug: 4.4.1
+      debug: 4.4.3
       espree: 9.6.1
       globals: 13.24.0
       ignore: 5.3.1
@@ -5758,7 +5790,7 @@ snapshots:
   '@humanwhocodes/config-array@0.11.14':
     dependencies:
       '@humanwhocodes/object-schema': 2.0.3
-      debug: 4.4.1
+      debug: 4.4.3
       minimatch: 3.1.2
     transitivePeerDependencies:
       - supports-color
@@ -6521,6 +6553,8 @@ snapshots:
     transitivePeerDependencies:
       - '@types/node'
 
+  '@speechify/api@2.0.0': {}
+
   '@standard-schema/spec@1.1.0': {}
 
   '@trivago/prettier-plugin-sort-imports@4.3.0(prettier@3.2.5)':
@@ -6647,7 +6681,7 @@ snapshots:
     dependencies:
       '@typescript-eslint/typescript-estree': 6.21.0(typescript@5.9.3)
       '@typescript-eslint/utils': 6.21.0(eslint@8.57.0)(typescript@5.9.3)
-      debug: 4.4.1
+      debug: 4.4.3
       eslint: 8.57.0
       ts-api-utils: 1.3.0(typescript@5.9.3)
     optionalDependencies:
@@ -6661,7 +6695,7 @@ snapshots:
     dependencies:
       '@typescript-eslint/types': 6.21.0
       '@typescript-eslint/visitor-keys': 6.21.0
-      debug: 4.4.1
+      debug: 4.4.3
       globby: 11.1.0
       is-glob: 4.0.3
       minimatch: 9.0.3
@@ -7360,7 +7394,7 @@ snapshots:
 
   eslint-import-resolver-typescript@3.6.1(@typescript-eslint/parser@6.21.0(eslint@8.57.0)(typescript@5.9.3))(eslint-import-resolver-node@0.3.9)(eslint-plugin-import@2.29.1)(eslint@8.57.0):
     dependencies:
-      debug: 4.4.1
+      debug: 4.4.3
       enhanced-resolve: 5.16.1
       eslint: 8.57.0
       eslint-module-utils: 2.8.1(@typescript-eslint/parser@6.21.0(eslint@8.57.0)(typescript@5.9.3))(eslint-import-resolver-node@0.3.9)(eslint-import-resolver-typescript@3.6.1(@typescript-eslint/parser@6.21.0(eslint@8.57.0)(typescript@5.9.3))(eslint-import-resolver-node@0.3.9)(eslint-plugin-import@2.29.1)(eslint@8.57.0))(eslint@8.57.0)
@@ -9077,7 +9111,7 @@ snapshots:
       cac: 6.7.14
       chokidar: 4.0.3
       consola: 3.4.2
-      debug: 4.4.1
+      debug: 4.4.3
       esbuild: 0.25.2
       joycon: 3.1.1
       picocolors: 1.1.1
@@ -9105,7 +9139,7 @@ snapshots:
       cac: 6.7.14
       chokidar: 4.0.3
       consola: 3.4.2
-      debug: 4.4.1
+      debug: 4.4.3
       esbuild: 0.25.2
       joycon: 3.1.1
       picocolors: 1.1.1
diff --git a/turbo.json b/turbo.json
index b3a1c44c5..b68291059 100644
--- a/turbo.json
+++ b/turbo.json
@@ -87,6 +87,7 @@
     "RUNWAY_AVATAR_PRESET_ID",
     "SARVAM_API_KEY",
     "SONIOX_API_KEY",
+    "SPEECHIFY_API_KEY",
     "SIP_PARTICIPANT_IDENTITY",
     "SIP_PHONE_NUMBER",
     "LK_OPENAI_DEBUG",

From d39adea4ac34ba30d21c1883c78ba6ea4bfc61dc Mon Sep 17 00:00:00 2001
From: luke-speechify <289678208+luke-speechify@users.noreply.github.com>
Date: Mon, 6 Jul 2026 13:36:15 +0100
Subject: [PATCH 02/13] chore: add changeset for speechify plugin

---
 .changeset/add-speechify-plugin.md | 5 +++++
 1 file changed, 5 insertions(+)
 create mode 100644 .changeset/add-speechify-plugin.md

diff --git a/.changeset/add-speechify-plugin.md b/.changeset/add-speechify-plugin.md
new file mode 100644
index 000000000..72ff73c75
--- /dev/null
+++ b/.changeset/add-speechify-plugin.md
@@ -0,0 +1,5 @@
+---
+"@livekit/agents-plugin-speechify": patch
+---
+
+feat(speechify): add Speechify TTS plugin with streaming and word-level timestamps

From 1c6a58139231b0d2337c98807e61670e4c301cae Mon Sep 17 00:00:00 2001
From: luke-speechify <289678208+luke-speechify@users.noreply.github.com>
Date: Mon, 6 Jul 2026 13:58:03 +0100
Subject: [PATCH 03/13] chore: use LiveKit copyright headers to match repo
 convention

---
 plugins/speechify/README.md     | 2 +-
 plugins/speechify/src/models.ts | 2 +-
 plugins/speechify/src/tts.ts    | 2 +-
 3 files changed, 3 insertions(+), 3 deletions(-)

diff --git a/plugins/speechify/README.md b/plugins/speechify/README.md
index 4fe5f6dcf..22827844c 100644
--- a/plugins/speechify/README.md
+++ b/plugins/speechify/README.md
@@ -1,5 +1,5 @@
 <!--
-SPDX-FileCopyrightText: 2025 Speechify, Inc.
+SPDX-FileCopyrightText: 2024 LiveKit, Inc.
 
 SPDX-License-Identifier: Apache-2.0
 -->
diff --git a/plugins/speechify/src/models.ts b/plugins/speechify/src/models.ts
index 8c4b54a69..f5b83dfc6 100644
--- a/plugins/speechify/src/models.ts
+++ b/plugins/speechify/src/models.ts
@@ -1,4 +1,4 @@
-// SPDX-FileCopyrightText: 2025 Speechify, Inc.
+// SPDX-FileCopyrightText: 2024 LiveKit, Inc.
 //
 // SPDX-License-Identifier: Apache-2.0
 
diff --git a/plugins/speechify/src/tts.ts b/plugins/speechify/src/tts.ts
index 032119e67..72abca5ec 100644
--- a/plugins/speechify/src/tts.ts
+++ b/plugins/speechify/src/tts.ts
@@ -1,4 +1,4 @@
-// SPDX-FileCopyrightText: 2025 Speechify, Inc.
+// SPDX-FileCopyrightText: 2024 LiveKit, Inc.
 //
 // SPDX-License-Identifier: Apache-2.0
 import {

From 26fede213d102a7aac5007c46fe32c4852dc5ded Mon Sep 17 00:00:00 2001
From: luke-speechify <289678208+luke-speechify@users.noreply.github.com>
Date: Mon, 6 Jul 2026 14:18:05 +0100
Subject: [PATCH 04/13] fix(speechify): flag final frame per segment so metrics
 are emitted

---
 plugins/speechify/src/tts.ts | 28 ++++++++++++++++++----------
 1 file changed, 18 insertions(+), 10 deletions(-)

diff --git a/plugins/speechify/src/tts.ts b/plugins/speechify/src/tts.ts
index 72abca5ec..eefbc7130 100644
--- a/plugins/speechify/src/tts.ts
+++ b/plugins/speechify/src/tts.ts
@@ -12,6 +12,7 @@ import {
   tokenize,
   tts,
 } from '@livekit/agents';
+import type { AudioFrame } from '@livekit/rtc-node';
 import type { Speechify } from '@speechify/api';
 import { SpeechifyClient, SpeechifyError } from '@speechify/api';
 import type { TTSModels } from './models.js';
@@ -243,19 +244,26 @@ export class SynthesizeStream extends tts.SynthesizeStream {
       const timed = timedStringsFromMarks(response.speech_marks, cumulativeDuration);
       let attached = false;
 
-      const putFrames = (frames: ReturnType<AudioByteStream['write']>) => {
-        for (const frame of frames) {
-          if (!attached && timed.length > 0) {
-            frame.userdata[USERDATA_TIMED_TRANSCRIPT] = timed;
-            attached = true;
-          }
-          cumulativeDuration += frame.samplesPerChannel / frame.sampleRate;
-          this.queue.put({ requestId, frame, final: false, segmentId: requestId });
+      // Defer emitting each frame by one so the last frame of this segment can
+      // be flagged final: true, which the base class needs to emit per-segment
+      // metrics (matches the elevenlabs/cartesia plugins).
+      let lastFrame: AudioFrame | undefined;
+      const sendFrame = (final: boolean) => {
+        if (!lastFrame) return;
+        if (!attached && timed.length > 0) {
+          lastFrame.userdata[USERDATA_TIMED_TRANSCRIPT] = timed;
+          attached = true;
         }
+        cumulativeDuration += lastFrame.samplesPerChannel / lastFrame.sampleRate;
+        this.queue.put({ requestId, frame: lastFrame, final, segmentId: requestId });
+        lastFrame = undefined;
       };
 
-      putFrames(bstream.write(audio));
-      putFrames(bstream.flush());
+      for (const frame of [...bstream.write(audio), ...bstream.flush()]) {
+        sendFrame(false);
+        lastFrame = frame;
+      }
+      sendFrame(true);
     };
 
     const consume = async () => {

From d1236468df0f2710caa884931f16c545e29023d0 Mon Sep 17 00:00:00 2001
From: luke-speechify <289678208+luke-speechify@users.noreply.github.com>
Date: Mon, 6 Jul 2026 14:43:36 +0100
Subject: [PATCH 05/13] feat(speechify): add simba-3.2 and default to
 dominic_32 voice

---
 plugins/speechify/src/models.ts |  2 +-
 plugins/speechify/src/tts.ts    | 11 +++++++++--
 2 files changed, 10 insertions(+), 3 deletions(-)

diff --git a/plugins/speechify/src/models.ts b/plugins/speechify/src/models.ts
index f5b83dfc6..69ef2f9ea 100644
--- a/plugins/speechify/src/models.ts
+++ b/plugins/speechify/src/models.ts
@@ -2,4 +2,4 @@
 //
 // SPDX-License-Identifier: Apache-2.0
 
-export type TTSModels = 'simba-english' | 'simba-multilingual' | 'simba-3.0';
+export type TTSModels = 'simba-english' | 'simba-multilingual' | 'simba-3.0' | 'simba-3.2';
diff --git a/plugins/speechify/src/tts.ts b/plugins/speechify/src/tts.ts
index eefbc7130..83511839b 100644
--- a/plugins/speechify/src/tts.ts
+++ b/plugins/speechify/src/tts.ts
@@ -20,7 +20,8 @@ import type { TTSModels } from './models.js';
 const NUM_CHANNELS = 1;
 const SAMPLE_RATE = 24000;
 const AUDIO_FORMAT: Speechify.GetSpeechRequest.AudioFormat = 'pcm';
-const DEFAULT_VOICE_ID = 'jack';
+const DEFAULT_VOICE_ID = 'dominic_32';
+const DEFAULT_MODEL: TTSModels = 'simba-3.2';
 
 export interface TTSOptions {
   voiceId: string;
@@ -36,6 +37,7 @@ export interface TTSOptions {
 
 const defaultOptions = (): Omit<TTSOptions, 'client' | 'tokenizer'> => ({
   voiceId: DEFAULT_VOICE_ID,
+  model: DEFAULT_MODEL,
 });
 
 const buildSpeechRequest = (text: string, opts: TTSOptions): Speechify.GetSpeechRequest => {
@@ -44,7 +46,9 @@ const buildSpeechRequest = (text: string, opts: TTSOptions): Speechify.GetSpeech
     input: text,
     voice_id: opts.voiceId,
   };
-  if (opts.model) request.model = opts.model;
+  // The SDK's Model union can lag newly released models; TTSModels is the
+  // plugin's source of truth and the API accepts the value.
+  if (opts.model) request.model = opts.model as Speechify.GetSpeechRequest.Model;
   if (opts.language) request.language = opts.language;
   if (opts.loudnessNormalization !== undefined || opts.textNormalization !== undefined) {
     request.options = {
@@ -80,6 +84,9 @@ export class TTS extends tts.TTS {
    * (24 kHz mono) plus word-level speech marks. `stream()` chunks input into
    * sentences and issues one request per sentence, emitting audio and aligned
    * word timestamps as each sentence completes.
+   *
+   * Defaults to the `dominic_32` voice and the `simba-3.2` model. The voice must
+   * support the chosen model; see the `/v1/voices` endpoint.
    */
   constructor(opts: Partial<TTSOptions> = {}) {
     const merged = { ...defaultOptions(), ...opts };

From 79a06ad9e9fed0f42a8e36c964002ef352426ff5 Mon Sep 17 00:00:00 2001
From: luke-speechify <289678208+luke-speechify@users.noreply.github.com>
Date: Mon, 6 Jul 2026 17:13:57 +0100
Subject: [PATCH 06/13] fix(speechify): pass through APIError; use 2026 license
 headers

---
 plugins/speechify/README.md     | 2 +-
 plugins/speechify/src/index.ts  | 2 +-
 plugins/speechify/src/models.ts | 2 +-
 plugins/speechify/src/tts.ts    | 6 +++++-
 4 files changed, 8 insertions(+), 4 deletions(-)

diff --git a/plugins/speechify/README.md b/plugins/speechify/README.md
index 22827844c..8468d7ceb 100644
--- a/plugins/speechify/README.md
+++ b/plugins/speechify/README.md
@@ -1,5 +1,5 @@
 <!--
-SPDX-FileCopyrightText: 2024 LiveKit, Inc.
+SPDX-FileCopyrightText: 2026 LiveKit, Inc.
 
 SPDX-License-Identifier: Apache-2.0
 -->
diff --git a/plugins/speechify/src/index.ts b/plugins/speechify/src/index.ts
index 252e901dd..cd06b682b 100644
--- a/plugins/speechify/src/index.ts
+++ b/plugins/speechify/src/index.ts
@@ -1,4 +1,4 @@
-// SPDX-FileCopyrightText: 2025 LiveKit, Inc.
+// SPDX-FileCopyrightText: 2026 LiveKit, Inc.
 //
 // SPDX-License-Identifier: Apache-2.0
 import { Plugin } from '@livekit/agents';
diff --git a/plugins/speechify/src/models.ts b/plugins/speechify/src/models.ts
index 69ef2f9ea..6d54a43e1 100644
--- a/plugins/speechify/src/models.ts
+++ b/plugins/speechify/src/models.ts
@@ -1,4 +1,4 @@
-// SPDX-FileCopyrightText: 2024 LiveKit, Inc.
+// SPDX-FileCopyrightText: 2026 LiveKit, Inc.
 //
 // SPDX-License-Identifier: Apache-2.0
 
diff --git a/plugins/speechify/src/tts.ts b/plugins/speechify/src/tts.ts
index 83511839b..1a66f4e8d 100644
--- a/plugins/speechify/src/tts.ts
+++ b/plugins/speechify/src/tts.ts
@@ -1,9 +1,10 @@
-// SPDX-FileCopyrightText: 2024 LiveKit, Inc.
+// SPDX-FileCopyrightText: 2026 LiveKit, Inc.
 //
 // SPDX-License-Identifier: Apache-2.0
 import {
   type APIConnectOptions,
   APIConnectionError,
+  APIError,
   APIStatusError,
   AudioByteStream,
   USERDATA_TIMED_TRANSCRIPT,
@@ -60,6 +61,9 @@ const buildSpeechRequest = (text: string, opts: TTSOptions): Speechify.GetSpeech
 };
 
 const toError = (e: unknown): Error => {
+  if (e instanceof APIError) {
+    return e;
+  }
   if (e instanceof SpeechifyError) {
     return new APIStatusError({ message: e.message, options: { statusCode: e.statusCode ?? -1 } });
   }

From aa53d11bef7c2210799632b8b3f0146611e32af7 Mon Sep 17 00:00:00 2001
From: luke-speechify <289678208+luke-speechify@users.noreply.github.com>
Date: Mon, 13 Jul 2026 22:09:41 +0100
Subject: [PATCH 07/13] fix(speechify): align plugin with @speechify/api v3.0.1
 (token= auth, drop model cast)

The SDK v3.0.0 renamed the auth constructor option from apiKey to token, and
v3.0.1's GetSpeechRequest.Model union now includes simba-3.0 and simba-3.2
natively so the plugin no longer needs to widen it via a cast.

- plugins/speechify/package.json: @speechify/api ^2.0.0 -> ^3.0.1
- plugins/speechify/src/tts.ts:
  - rename public TTSOptions.apiKey -> token; pass through to
    new SpeechifyClient({ token, baseUrl })
  - update JSDoc + local variable + Omit-union for updateOptions
  - drop the `opts.model as Speechify.GetSpeechRequest.Model` cast + comment;
    v3.0.1's Model union already covers simba-english / simba-multilingual /
    simba-3.0 / simba-3.2
- pnpm-lock.yaml: refreshed
---
 plugins/speechify/package.json |  2 +-
 plugins/speechify/src/tts.ts   | 18 ++++++++----------
 pnpm-lock.yaml                 | 10 +++++-----
 3 files changed, 14 insertions(+), 16 deletions(-)

diff --git a/plugins/speechify/package.json b/plugins/speechify/package.json
index fbf45fcac..fe7bd5514 100644
--- a/plugins/speechify/package.json
+++ b/plugins/speechify/package.json
@@ -43,7 +43,7 @@
     "typescript": "^5.0.0"
   },
   "dependencies": {
-    "@speechify/api": "^2.0.0"
+    "@speechify/api": "^3.0.1"
   },
   "peerDependencies": {
     "@livekit/agents": "workspace:*",
diff --git a/plugins/speechify/src/tts.ts b/plugins/speechify/src/tts.ts
index 1a66f4e8d..c0e16a235 100644
--- a/plugins/speechify/src/tts.ts
+++ b/plugins/speechify/src/tts.ts
@@ -30,7 +30,7 @@ export interface TTSOptions {
   language?: string;
   loudnessNormalization?: boolean;
   textNormalization?: boolean;
-  apiKey?: string;
+  token?: string;
   baseUrl?: string;
   client?: SpeechifyClient;
   tokenizer?: tokenize.SentenceTokenizer;
@@ -47,9 +47,7 @@ const buildSpeechRequest = (text: string, opts: TTSOptions): Speechify.GetSpeech
     input: text,
     voice_id: opts.voiceId,
   };
-  // The SDK's Model union can lag newly released models; TTSModels is the
-  // plugin's source of truth and the API accepts the value.
-  if (opts.model) request.model = opts.model as Speechify.GetSpeechRequest.Model;
+  if (opts.model) request.model = opts.model;
   if (opts.language) request.language = opts.language;
   if (opts.loudnessNormalization !== undefined || opts.textNormalization !== undefined) {
     request.options = {
@@ -80,9 +78,9 @@ export class TTS extends tts.TTS {
    * Create a new instance of Speechify TTS.
    *
    * @remarks
-   * `apiKey` must be set, either via the constructor or the `SPEECHIFY_API_KEY`
+   * `token` must be set, either via the constructor or the `SPEECHIFY_API_KEY`
    * environment variable. Pass a preconfigured `client` to reuse an existing
-   * `SpeechifyClient` (in which case `apiKey`/`baseUrl` are ignored).
+   * `SpeechifyClient` (in which case `token`/`baseUrl` are ignored).
    *
    * Synthesis uses the Speechify `/audio/speech` endpoint, which returns raw PCM
    * (24 kHz mono) plus word-level speech marks. `stream()` chunks input into
@@ -103,13 +101,13 @@ export class TTS extends tts.TTS {
     if (merged.client) {
       this.#client = merged.client;
     } else {
-      const apiKey = merged.apiKey ?? process.env.SPEECHIFY_API_KEY;
-      if (!apiKey) {
+      const token = merged.token ?? process.env.SPEECHIFY_API_KEY;
+      if (!token) {
         throw new Error(
           'Speechify API key is required, whether as an argument or as $SPEECHIFY_API_KEY',
         );
       }
-      this.#client = new SpeechifyClient({ apiKey, baseUrl: merged.baseUrl });
+      this.#client = new SpeechifyClient({ token, baseUrl: merged.baseUrl });
     }
   }
 
@@ -133,7 +131,7 @@ export class TTS extends tts.TTS {
     return this.#tokenizer;
   }
 
-  updateOptions(opts: Partial<Omit<TTSOptions, 'client' | 'apiKey' | 'baseUrl' | 'tokenizer'>>) {
+  updateOptions(opts: Partial<Omit<TTSOptions, 'client' | 'token' | 'baseUrl' | 'tokenizer'>>) {
     this.#opts = { ...this.#opts, ...opts };
   }
 
diff --git a/pnpm-lock.yaml b/pnpm-lock.yaml
index 10262223b..9b5078b9f 100644
--- a/pnpm-lock.yaml
+++ b/pnpm-lock.yaml
@@ -1330,8 +1330,8 @@ importers:
   plugins/speechify:
     dependencies:
       '@speechify/api':
-        specifier: ^2.0.0
-        version: 2.0.0
+        specifier: ^3.0.1
+        version: 3.0.1
     devDependencies:
       '@livekit/agents':
         specifier: workspace:*
@@ -2741,8 +2741,8 @@ packages:
   '@rushstack/ts-command-line@4.21.0':
     resolution: {integrity: sha512-z38FLUCn8M9FQf19gJ9eltdwkvc47PxvJmVZS6aKwbBAa3Pis3r3A+ZcBCVPNb9h/Tbga+i0tHdzoSGUoji9GQ==}
 
-  '@speechify/api@2.0.0':
-    resolution: {integrity: sha512-GYIYO/B0bRArg9e4ljLxHL/ZG27p7X68UGu6bvEqBk0I8CSpqPKHO+rsGGJR+oIRzXZpR5LVNcS6RPN0ljnvNw==}
+  '@speechify/api@3.0.1':
+    resolution: {integrity: sha512-bEqbyvwGW0A/cgq+ZsQJVP3Hm4/0Bm1sniVtPLBh7vDNG2f/YY8y8sweWojoUnhIeIAL6bHBde43+VZLA2MYpA==}
     engines: {node: '>=18.0.0'}
 
   '@standard-schema/spec@1.1.0':
@@ -6553,7 +6553,7 @@ snapshots:
     transitivePeerDependencies:
       - '@types/node'
 
-  '@speechify/api@2.0.0': {}
+  '@speechify/api@3.0.1': {}
 
   '@standard-schema/spec@1.1.0': {}
 

From 9951b068075cd20a9692af6da1c22fc4904bc06f Mon Sep 17 00:00:00 2001
From: luke-speechify <289678208+luke-speechify@users.noreply.github.com>
Date: Mon, 13 Jul 2026 22:30:56 +0100
Subject: [PATCH 08/13] fix(speechify): call markStarted before each synthesis
 request

Addresses Devin review feedback on livekit/agents-js#1968.

The base class SynthesizeStream.monitorMetrics gates all TTS metric
emission on this.#startedHrTime being set, which only happens via
markStarted(). Without it, TTFB, duration, and character-count metrics
were silently dropped for all Speechify streaming synthesis, and
USERDATA_TTS_STARTED_TIME never got stamped on frames so downstream
transcription timing was degraded.

Matches the elevenlabs / cartesia / resemble / deepgram plugin pattern.
---
 plugins/speechify/src/tts.ts | 1 +
 1 file changed, 1 insertion(+)

diff --git a/plugins/speechify/src/tts.ts b/plugins/speechify/src/tts.ts
index c0e16a235..a7f742f0c 100644
--- a/plugins/speechify/src/tts.ts
+++ b/plugins/speechify/src/tts.ts
@@ -244,6 +244,7 @@ export class SynthesizeStream extends tts.SynthesizeStream {
       const requestId = shortuuid();
       const bstream = new AudioByteStream(SAMPLE_RATE, NUM_CHANNELS);
 
+      this.markStarted();
       const response = await this.#tts.client.audio.speech(buildSpeechRequest(text, this.#opts), {
         abortSignal: this.abortSignal,
         timeoutInSeconds: this.connOptions.timeoutMs / 1000,

From 8e0669e6175d32bf1482e969667c9e499720a656 Mon Sep 17 00:00:00 2001
From: luke-speechify <289678208+luke-speechify@users.noreply.github.com>
Date: Mon, 13 Jul 2026 22:39:09 +0100
Subject: [PATCH 09/13] fix(speechify): stop closing output queue in run()
 error paths

Addresses Devin review findings on livekit/agents-js#1968 that both
Chunked and Synthesize streams close `this.queue` in their run()
catch blocks before re-throwing. The base class already closes the
queue in its own `finally` after run() returns, and it retries run()
on retryable APIErrors - so closing on error made the next retry's
`queue.put(...)` throw 'Queue is closed', turning transient network
errors into permanent failures.

Matches the guard the minimax plugin documents on its own catch.
---
 plugins/speechify/src/tts.ts | 2 --
 1 file changed, 2 deletions(-)

diff --git a/plugins/speechify/src/tts.ts b/plugins/speechify/src/tts.ts
index a7f742f0c..c9d056cf4 100644
--- a/plugins/speechify/src/tts.ts
+++ b/plugins/speechify/src/tts.ts
@@ -205,7 +205,6 @@ export class ChunkedStream extends tts.ChunkedStream {
       putFrames(bstream.flush());
       this.queue.close();
     } catch (e) {
-      if (!this.queue.closed) this.queue.close();
       if (this.abortSignal.aborted) return;
       throw toError(e);
     }
@@ -289,7 +288,6 @@ export class SynthesizeStream extends tts.SynthesizeStream {
     try {
       await Promise.all([forwardInput(), consume()]);
     } catch (e) {
-      if (!this.queue.closed) this.queue.close();
       if (this.abortSignal.aborted) return;
       throw toError(e);
     }

From d86166598b7a10b4849a1bafdc594310b9bb91e6 Mon Sep 17 00:00:00 2001
From: luke-speechify <289678208+luke-speechify@users.noreply.github.com>
Date: Mon, 13 Jul 2026 22:48:42 +0100
Subject: [PATCH 10/13] fix(speechify): keep sentence tokenizer alive after
 endInput

Addresses Devin review finding on livekit/agents-js#1968: calling
`sentenceStream.close()` immediately after `endInput()` closes the
tokenizer's output queue before its internal async loop has a chance
to flush the last buffered sentence, so the final `queue.put(token)`
throws 'Queue is closed' and consume() never sees the tail. Result:
the last sentence of a TTS response could be silently dropped.

Matches the neuphonic plugin pattern: only call endInput(), let the
tokenizer close its own output queue when its input drains.
---
 plugins/speechify/src/tts.ts | 1 -
 1 file changed, 1 deletion(-)

diff --git a/plugins/speechify/src/tts.ts b/plugins/speechify/src/tts.ts
index c9d056cf4..2d5fd83de 100644
--- a/plugins/speechify/src/tts.ts
+++ b/plugins/speechify/src/tts.ts
@@ -236,7 +236,6 @@ export class SynthesizeStream extends tts.SynthesizeStream {
         }
       }
       sentenceStream.endInput();
-      sentenceStream.close();
     };
 
     const synthesizeSentence = async (text: string) => {

From 987d28b61e7f3da9e38855ffde61cdea18658c71 Mon Sep 17 00:00:00 2001
From: luke-speechify <289678208+luke-speechify@users.noreply.github.com>
Date: Wed, 22 Jul 2026 20:38:33 +0100
Subject: [PATCH 11/13] fix(speechify): rename token to apiKey, hide SDK client
 from public API

- align auth arg with sibling plugins (was `token`, now `apiKey`)
- drop `client?: SpeechifyClient` option and `get client()` getter
- move the SDK call into an `@internal` `_synthesize` on TTS that returns
  `{ audio, timed }` in LiveKit-owned types
- revert `package.json` author to LiveKit
- drop "maintained by Speechify" paragraph from the plugin README

verified `dist/tts.d.ts` no longer references any `Speechify*` SDK types.
---
 plugins/speechify/README.md    |  4 ---
 plugins/speechify/package.json |  2 +-
 plugins/speechify/src/tts.ts   | 62 ++++++++++++++++------------------
 3 files changed, 31 insertions(+), 37 deletions(-)

diff --git a/plugins/speechify/README.md b/plugins/speechify/README.md
index 8468d7ceb..a4f60502a 100644
--- a/plugins/speechify/README.md
+++ b/plugins/speechify/README.md
@@ -15,7 +15,3 @@ for information on how to use it, or browse the [API
 reference](https://docs.livekit.io/agents-js/modules/plugins_agents_plugin_speechify.html).
 See the [repository](https://github.com/livekit/agents-js) for more information
 about the framework as a whole.
-
-This plugin is maintained by Speechify. Issues and contributions specific to the
-plugin are welcome at
-[Speechify-AI/livekit-plugin-js](https://github.com/Speechify-AI/livekit-plugin-js).
diff --git a/plugins/speechify/package.json b/plugins/speechify/package.json
index fe7bd5514..aab77e5f1 100644
--- a/plugins/speechify/package.json
+++ b/plugins/speechify/package.json
@@ -15,7 +15,7 @@
       "default": "./dist/index.cjs"
     }
   },
-  "author": "Speechify",
+  "author": "LiveKit",
   "type": "module",
   "repository": "git@github.com:livekit/agents-js.git",
   "license": "Apache-2.0",
diff --git a/plugins/speechify/src/tts.ts b/plugins/speechify/src/tts.ts
index 2d5fd83de..e47cf9cca 100644
--- a/plugins/speechify/src/tts.ts
+++ b/plugins/speechify/src/tts.ts
@@ -30,13 +30,12 @@ export interface TTSOptions {
   language?: string;
   loudnessNormalization?: boolean;
   textNormalization?: boolean;
-  token?: string;
+  apiKey?: string;
   baseUrl?: string;
-  client?: SpeechifyClient;
   tokenizer?: tokenize.SentenceTokenizer;
 }
 
-const defaultOptions = (): Omit<TTSOptions, 'client' | 'tokenizer'> => ({
+const defaultOptions = (): Omit<TTSOptions, 'tokenizer'> => ({
   voiceId: DEFAULT_VOICE_ID,
   model: DEFAULT_MODEL,
 });
@@ -78,9 +77,8 @@ export class TTS extends tts.TTS {
    * Create a new instance of Speechify TTS.
    *
    * @remarks
-   * `token` must be set, either via the constructor or the `SPEECHIFY_API_KEY`
-   * environment variable. Pass a preconfigured `client` to reuse an existing
-   * `SpeechifyClient` (in which case `token`/`baseUrl` are ignored).
+   * `apiKey` must be set, either via the constructor or the `SPEECHIFY_API_KEY`
+   * environment variable.
    *
    * Synthesis uses the Speechify `/audio/speech` endpoint, which returns raw PCM
    * (24 kHz mono) plus word-level speech marks. `stream()` chunks input into
@@ -98,17 +96,13 @@ export class TTS extends tts.TTS {
     this.#opts = merged;
     this.#tokenizer = merged.tokenizer ?? new tokenize.basic.SentenceTokenizer();
 
-    if (merged.client) {
-      this.#client = merged.client;
-    } else {
-      const token = merged.token ?? process.env.SPEECHIFY_API_KEY;
-      if (!token) {
-        throw new Error(
-          'Speechify API key is required, whether as an argument or as $SPEECHIFY_API_KEY',
-        );
-      }
-      this.#client = new SpeechifyClient({ token, baseUrl: merged.baseUrl });
+    const token = merged.apiKey ?? process.env.SPEECHIFY_API_KEY;
+    if (!token) {
+      throw new Error(
+        'Speechify API key is required, whether as an argument or as $SPEECHIFY_API_KEY',
+      );
     }
+    this.#client = new SpeechifyClient({ token, baseUrl: merged.baseUrl });
   }
 
   get model(): string {
@@ -119,10 +113,6 @@ export class TTS extends tts.TTS {
     return 'Speechify';
   }
 
-  get client(): SpeechifyClient {
-    return this.#client;
-  }
-
   get options(): TTSOptions {
     return this.#opts;
   }
@@ -131,7 +121,21 @@ export class TTS extends tts.TTS {
     return this.#tokenizer;
   }
 
-  updateOptions(opts: Partial<Omit<TTSOptions, 'client' | 'token' | 'baseUrl' | 'tokenizer'>>) {
+  /** @internal */
+  async _synthesize(
+    text: string,
+    opts: TTSOptions,
+    offsetSeconds: number,
+    params: { abortSignal: AbortSignal; timeoutInSeconds?: number },
+  ): Promise<{ audio: Buffer; timed: ReturnType<typeof createTimedString>[] }> {
+    const response = await this.#client.audio.speech(buildSpeechRequest(text, opts), params);
+    return {
+      audio: Buffer.from(response.audio_data, 'base64'),
+      timed: timedStringsFromMarks(response.speech_marks, offsetSeconds),
+    };
+  }
+
+  updateOptions(opts: Partial<Omit<TTSOptions, 'apiKey' | 'baseUrl' | 'tokenizer'>>) {
     this.#opts = { ...this.#opts, ...opts };
   }
 
@@ -182,13 +186,10 @@ export class ChunkedStream extends tts.ChunkedStream {
     const bstream = new AudioByteStream(SAMPLE_RATE, NUM_CHANNELS);
 
     try {
-      const response = await this.#tts.client.audio.speech(
-        buildSpeechRequest(this.inputText, this.#opts),
-        { abortSignal: this.abortSignal, timeoutInSeconds: this.#timeoutInSeconds },
-      );
-
-      const audio = Buffer.from(response.audio_data, 'base64');
-      const timed = timedStringsFromMarks(response.speech_marks, 0);
+      const { audio, timed } = await this.#tts._synthesize(this.inputText, this.#opts, 0, {
+        abortSignal: this.abortSignal,
+        timeoutInSeconds: this.#timeoutInSeconds,
+      });
       let attached = false;
 
       const putFrames = (frames: ReturnType<AudioByteStream['write']>) => {
@@ -243,13 +244,10 @@ export class SynthesizeStream extends tts.SynthesizeStream {
       const bstream = new AudioByteStream(SAMPLE_RATE, NUM_CHANNELS);
 
       this.markStarted();
-      const response = await this.#tts.client.audio.speech(buildSpeechRequest(text, this.#opts), {
+      const { audio, timed } = await this.#tts._synthesize(text, this.#opts, cumulativeDuration, {
         abortSignal: this.abortSignal,
         timeoutInSeconds: this.connOptions.timeoutMs / 1000,
       });
-
-      const audio = Buffer.from(response.audio_data, 'base64');
-      const timed = timedStringsFromMarks(response.speech_marks, cumulativeDuration);
       let attached = false;
 
       // Defer emitting each frame by one so the last frame of this segment can

From 30b3757819f0d9069e5663f162abac40be247127 Mon Sep 17 00:00:00 2001
From: luke-speechify <289678208+luke-speechify@users.noreply.github.com>
Date: Wed, 22 Jul 2026 20:53:24 +0100
Subject: [PATCH 12/13] fix(speechify): forward abortSignal and emit final:true
 once per reply

Devin re-review (#1968):

- TTS.synthesize and ChunkedStream now accept + forward `abortSignal` to
  super() so external cancels reach the one-shot request path (matches
  the abstract signature in agents/src/tts/tts.ts and the cartesia plugin)
- SynthesizeStream now emits `final: true` exactly once, on the last
  frame of the whole reply (previously per-sentence, which dropped
  sentences 2..N from the base-class metrics because flush() only
  records one pending metric text per reply). Matches elevenlabs/cartesia.
---
 plugins/speechify/src/tts.ts | 59 +++++++++++++++++++++++-------------
 1 file changed, 38 insertions(+), 21 deletions(-)

diff --git a/plugins/speechify/src/tts.ts b/plugins/speechify/src/tts.ts
index e47cf9cca..bd165b799 100644
--- a/plugins/speechify/src/tts.ts
+++ b/plugins/speechify/src/tts.ts
@@ -139,8 +139,12 @@ export class TTS extends tts.TTS {
     this.#opts = { ...this.#opts, ...opts };
   }
 
-  synthesize(text: string, connOptions?: APIConnectOptions): tts.ChunkedStream {
-    return new ChunkedStream(this, text, this.#opts, connOptions);
+  synthesize(
+    text: string,
+    connOptions?: APIConnectOptions,
+    abortSignal?: AbortSignal,
+  ): tts.ChunkedStream {
+    return new ChunkedStream(this, text, this.#opts, connOptions, abortSignal);
   }
 
   stream(options?: { connOptions?: APIConnectOptions }): tts.SynthesizeStream {
@@ -173,8 +177,14 @@ export class ChunkedStream extends tts.ChunkedStream {
   #tts: TTS;
   #timeoutInSeconds?: number;
 
-  constructor(ttsInstance: TTS, text: string, opts: TTSOptions, connOptions?: APIConnectOptions) {
-    super(text, ttsInstance, connOptions);
+  constructor(
+    ttsInstance: TTS,
+    text: string,
+    opts: TTSOptions,
+    connOptions?: APIConnectOptions,
+    abortSignal?: AbortSignal,
+  ) {
+    super(text, ttsInstance, connOptions, abortSignal);
     this.#tts = ttsInstance;
     this.#opts = opts;
     this.#timeoutInSeconds =
@@ -227,6 +237,23 @@ export class SynthesizeStream extends tts.SynthesizeStream {
     const sentenceStream = this.#tts.tokenizer.stream();
     let cumulativeDuration = 0;
 
+    // Buffer-one deferral across the WHOLE run: the base class only records
+    // one pending text per reply, so `final: true` must be emitted exactly
+    // once on the last frame of the entire stream (mirrors elevenlabs/cartesia).
+    let lastFrame: AudioFrame | undefined;
+    let lastRequestId: string | undefined;
+    const sendFrame = (final: boolean) => {
+      if (!lastFrame || !lastRequestId) return;
+      cumulativeDuration += lastFrame.samplesPerChannel / lastFrame.sampleRate;
+      this.queue.put({
+        requestId: lastRequestId,
+        frame: lastFrame,
+        final,
+        segmentId: lastRequestId,
+      });
+      lastFrame = undefined;
+    };
+
     const forwardInput = async () => {
       for await (const input of this.input) {
         if (this.abortController.signal.aborted) break;
@@ -248,28 +275,17 @@ export class SynthesizeStream extends tts.SynthesizeStream {
         abortSignal: this.abortSignal,
         timeoutInSeconds: this.connOptions.timeoutMs / 1000,
       });
-      let attached = false;
 
-      // Defer emitting each frame by one so the last frame of this segment can
-      // be flagged final: true, which the base class needs to emit per-segment
-      // metrics (matches the elevenlabs/cartesia plugins).
-      let lastFrame: AudioFrame | undefined;
-      const sendFrame = (final: boolean) => {
-        if (!lastFrame) return;
-        if (!attached && timed.length > 0) {
-          lastFrame.userdata[USERDATA_TIMED_TRANSCRIPT] = timed;
-          attached = true;
-        }
-        cumulativeDuration += lastFrame.samplesPerChannel / lastFrame.sampleRate;
-        this.queue.put({ requestId, frame: lastFrame, final, segmentId: requestId });
-        lastFrame = undefined;
-      };
+      const frames = [...bstream.write(audio), ...bstream.flush()];
+      if (timed.length > 0 && frames.length > 0) {
+        frames[0]!.userdata[USERDATA_TIMED_TRANSCRIPT] = timed;
+      }
 
-      for (const frame of [...bstream.write(audio), ...bstream.flush()]) {
+      for (const frame of frames) {
         sendFrame(false);
         lastFrame = frame;
+        lastRequestId = requestId;
       }
-      sendFrame(true);
     };
 
     const consume = async () => {
@@ -279,6 +295,7 @@ export class SynthesizeStream extends tts.SynthesizeStream {
         if (!text) continue;
         await synthesizeSentence(text);
       }
+      sendFrame(true);
       this.queue.put(SynthesizeStream.END_OF_STREAM);
     };
 

From cb85f9499ef41e26663699a5c16692199404f1ee Mon Sep 17 00:00:00 2001
From: luke-speechify <289678208+luke-speechify@users.noreply.github.com>
Date: Wed, 22 Jul 2026 21:12:14 +0100
Subject: [PATCH 13/13] fix(speechify): advance word-timestamp offset by full
 sentence duration

The buffer-one deferral in the previous commit meant `cumulativeDuration`
lagged one frame behind actual produced audio, so every sentence after the
first got a slightly early offset for its word marks (Devin re-review).

Track the total emitted duration independently: advance `offsetSeconds` by
the entire sentence's frame duration right after synthesis, before the
deferral loop. The deferral still governs when `final: true` fires, but
timestamps no longer share state with it.
---
 plugins/speechify/src/tts.ts | 10 +++++++---
 1 file changed, 7 insertions(+), 3 deletions(-)

diff --git a/plugins/speechify/src/tts.ts b/plugins/speechify/src/tts.ts
index bd165b799..1043a5b6c 100644
--- a/plugins/speechify/src/tts.ts
+++ b/plugins/speechify/src/tts.ts
@@ -235,7 +235,11 @@ export class SynthesizeStream extends tts.SynthesizeStream {
 
   protected async run() {
     const sentenceStream = this.#tts.tokenizer.stream();
-    let cumulativeDuration = 0;
+    // Total audio duration produced so far, used as the timestamp offset for
+    // the next sentence's word marks. Advances by each sentence's full frame
+    // duration right after synthesis, so it stays accurate even though the
+    // last frame of each sentence is deferred by the buffer-one loop below.
+    let offsetSeconds = 0;
 
     // Buffer-one deferral across the WHOLE run: the base class only records
     // one pending text per reply, so `final: true` must be emitted exactly
@@ -244,7 +248,6 @@ export class SynthesizeStream extends tts.SynthesizeStream {
     let lastRequestId: string | undefined;
     const sendFrame = (final: boolean) => {
       if (!lastFrame || !lastRequestId) return;
-      cumulativeDuration += lastFrame.samplesPerChannel / lastFrame.sampleRate;
       this.queue.put({
         requestId: lastRequestId,
         frame: lastFrame,
@@ -271,7 +274,7 @@ export class SynthesizeStream extends tts.SynthesizeStream {
       const bstream = new AudioByteStream(SAMPLE_RATE, NUM_CHANNELS);
 
       this.markStarted();
-      const { audio, timed } = await this.#tts._synthesize(text, this.#opts, cumulativeDuration, {
+      const { audio, timed } = await this.#tts._synthesize(text, this.#opts, offsetSeconds, {
         abortSignal: this.abortSignal,
         timeoutInSeconds: this.connOptions.timeoutMs / 1000,
       });
@@ -280,6 +283,7 @@ export class SynthesizeStream extends tts.SynthesizeStream {
       if (timed.length > 0 && frames.length > 0) {
         frames[0]!.userdata[USERDATA_TIMED_TRANSCRIPT] = timed;
       }
+      offsetSeconds += frames.reduce((sum, f) => sum + f.samplesPerChannel / f.sampleRate, 0);
 
       for (const frame of frames) {
         sendFrame(false);