TurboWarp-Web-Speech

English | 日本語

A TurboWarp extension for speech recognition and speech synthesis in the browser, using the Web Speech API. It needs no API key and no server of its own.

User guide: English

What it does

• Listens to the microphone and reports recognized text, once or continuously.
• Starts hats when speech is recognized and when someone starts speaking, so a project can stop speech output when the user interrupts.
• Speaks text with the browser's voices, with a choice of language, voice, rate, and pitch.
• Provides a Composition API so other extensions (such as a voice-chat extension) can use the same capability with their own blocks.

Requirements and safety

• TurboWarp Web in a browser with the Web Speech API. Speech recognition works in Chrome, Edge, and Safari; Firefox does not provide it. Speech synthesis works in most browsers; available voices depend on the OS.
• TurboWarp Desktop: speech synthesis works, but speech recognition usually fails because Electron does not include the recognition service.

[!IMPORTANT] This extension must run unsandboxed. Sandboxed extensions run in a worker, which has no speech APIs, and the extension starts hats through the VM runtime. Load extensions only from sources you trust.

• Privacy: Chrome and Edge normally send microphone audio to the browser vendor's cloud service for recognition. Explain this to users, especially in schools, before using speech recognition.
• The browser asks for microphone permission the first time listening starts.

Installation

Built JavaScript

1. Download dist/turbowarp-web-speech.js.
2. Open Extensions in TurboWarp.
3. Choose Custom Extension and load the file.
4. Enable Run without sandbox.

npm package

pnpm add --save-exact @kubohiroya/[email protected]

Standalone bundle:

node_modules/@kubohiroya/turbowarp-web-speech/dist/turbowarp-web-speech.js

Quick start

when green flag clicked
set recognition language to [ja-JP]
set speech language to [ja-JP]
say (listen and wait)
speak (join [you said ] (recognized text)) and wait

when someone starts speaking
stop speaking

Block reference

The block reference is generated from src/block-definitions.json. Do not edit the generated section manually.

speech recognition supported?

Reports whether this browser offers speech recognition.

| Property | Value | |---|---| | Type | Boolean | | Opcode | isRecognitionSupported |

set recognition language to [LANG]

Sets the language used from the next time listening starts.

| Property | Value | |---|---| | Type | Command | | Opcode | setRecognitionLanguage | | LANG | String, default: ja-JP, choices: ja-JP, en-US, en-GB, zh-CN, ko-KR, fr-FR, de-DE, es-ES |

start listening [MODE]

Starts listening. once stops after one utterance; continuous keeps listening until stop listening.

| Property | Value | |---|---| | Type | Command | | Opcode | startListening | | MODE | String, default: once, choices: once, continuous |

stop listening

Stops listening and delivers any pending result.

| Property | Value | |---|---| | Type | Command | | Opcode | stopListening |

listening?

Reports whether the microphone is being listened to.

| Property | Value | |---|---| | Type | Boolean | | Opcode | isListening |

listen and wait

Listens for one utterance and reports its text, or empty text when nothing was recognized.

| Property | Value | |---|---| | Type | Reporter | | Opcode | listenAndWait |

when speech is recognized

Starts each time an utterance is recognized.

| Property | Value | |---|---| | Type | Hat | | Opcode | whenSpeechRecognized |

when someone starts speaking

Starts when the recognizer detects the start of speech. Useful for stopping speech output when the user interrupts.

| Property | Value | |---|---| | Type | Hat | | Opcode | whenSpeechStarts |

recognized text

Reports the most recently recognized utterance.

| Property | Value | |---|---| | Type | Reporter | | Opcode | recognizedText |

interim recognized text

Reports the text recognized so far for the utterance in progress.

| Property | Value | |---|---| | Type | Reporter | | Opcode | interimText |

speech synthesis supported?

Reports whether this browser offers speech synthesis.

| Property | Value | |---|---| | Type | Boolean | | Opcode | isSynthesisSupported |

speak [TEXT]

Starts speaking the text and continues the script immediately.

| Property | Value | |---|---| | Type | Command | | Opcode | speak | | TEXT | String, default: こんにちは |

speak [TEXT] and wait

Speaks the text and waits until it ends or is stopped.

| Property | Value | |---|---| | Type | Command | | Opcode | speakAndWait | | TEXT | String, default: こんにちは |

stop speaking

Stops speech output, including queued text.

| Property | Value | |---|---| | Type | Command | | Opcode | stopSpeaking |

speaking?

Reports whether speech output is in progress.

| Property | Value | |---|---| | Type | Boolean | | Opcode | isSpeaking |

set speech language to [LANG]

Sets the language of speech output.

| Property | Value | |---|---| | Type | Command | | Opcode | setSpeechLanguage | | LANG | String, default: ja-JP, choices: ja-JP, en-US, en-GB, zh-CN, ko-KR, fr-FR, de-DE, es-ES |

set speech voice to [VOICE]

Chooses a voice by name. Available voices depend on the browser and OS; empty uses the default voice for the language.

| Property | Value | |---|---| | Type | Command | | Opcode | setSpeechVoice | | VOICE | String, default: ``, choices: read from the browser at run time |

set speech rate to [RATE]

Sets the speaking rate (0.1 to 10, normal 1).

| Property | Value | |---|---| | Type | Command | | Opcode | setSpeechRate | | RATE | Number, default: 1 |

set speech pitch to [PITCH]

Sets the speaking pitch (0 to 2, normal 1).

| Property | Value | |---|---| | Type | Command | | Opcode | setSpeechPitch | | PITCH | Number, default: 1 |

available voices

Reports the voices of this browser as JSON text.

| Property | Value | |---|---| | Type | Reporter | | Opcode | voicesJson |

last speech error

Reports the most recent recognition or synthesis error, or an empty string.

| Property | Value | |---|---| | Type | Reporter | | Opcode | lastError |

Important behavior

| Situation | Behavior | |---|---| | start listening [once] | Stops after one utterance. | | start listening [continuous] | Keeps listening until stop listening. The browser ends sessions after silence; the extension restarts them automatically. | | No speech detected | Reported in last speech error; continuous mode keeps listening. | | Microphone denied, no microphone, or the service unreachable | Listening stops, even in continuous mode, and last speech error explains why. | | listen and wait | Reports the first recognized utterance, or empty text when nothing was recognized. | | speak while speaking | The new text is queued after the current one. | | stop speaking | Stops the current and queued text; speak and wait then continues without an error. | | Unknown voice name | The default voice for the speech language is used. | | Project stop | Listening stops immediately (the microphone is released) and speech is silenced. |

Composition API

Importing the Composition API does not register the standalone TurboWarp extension.

import {createWebSpeech} from '@kubohiroya/turbowarp-web-speech/composition';

const speech = createWebSpeech();
speech.setRecognitionLanguage('ja-JP');
speech.configureSpeech({lang: 'ja-JP', rate: 1.1});

speech.subscribe((event) => {
  if (event.type === 'speechStart') speech.cancelSpeech(); // barge-in
  if (event.type === 'final') console.log(event.text);
});

speech.startListening('continuous');
await speech.speak('こんにちは');

speech.release();

| Member | Purpose | |---|---| | isRecognitionSupported / setRecognitionLanguage / startListening / stopListening / abortListening / listening | Recognition control | | listenOnce() | Resolve with one utterance, or '' | | lastRecognizedText / interimText / lastConfidence | Latest results | | isSynthesisSupported / configureSpeech / speechSettings / voices / speak / cancelSpeech / speaking | Synthesis control; speak resolves when the utterance ends or is cancelled | | subscribe | Events: listening, speechStart, interim, final, error, speaking | | release | Stop listening and speaking, detach listeners |

Pass environment to createWebSpeech to substitute the browser APIs, for example in tests.

Compatibility

| Identifier | Value | Stability | |---|---|---| | Product name | TurboWarp-Web-Speech | Human-facing | | Repository | kubohiroya/turbowarp-web-speech | Current source location | | npm package | @kubohiroya/turbowarp-web-speech | Public package contract | | Extension ID | kubohiroyawebspeech | Stored in SB3; migration required to change | | Composition API | @kubohiroya/turbowarp-web-speech/composition | Public package contract |

Development

Use Node.js 22.18.0 or newer and the pnpm version declared by packageManager.

corepack enable
pnpm install --frozen-lockfile
pnpm check

License

Mozilla Public License 2.0 (SPDX: MPL-2.0).