內容簡介
總結文案：Speechable 是一個 WordPress 外掛，利用 AI 技術將您的內容轉換成自然聲音的音頻。該外掛使用 Piper TTS，一個在瀏覽器中運行的開源神經文本轉語音引擎。
問題與答案：
1. Speechable 外掛使用了哪些第三方服務來實現文字轉語音功能？
- Piper TTS Web Library（由 Mintplex Labs 提供）
- Whisper（由 Hugging Face Transformers.js 提供）
2. 在使用 Speechable 外掛時，哪些資料會被傳輸？
- 在生成音頻時，TTS程式庫和ONNX運行時庫會從CDN加載
- 當首次使用某種語音時，語音模型文件將從Hugging Face下載
- Whisper模型僅在第一次使用時用於單詞時間戳提取
- 您的內容文本在本地瀏覽器中進行處理，不會發送到任何外部服務器
3. 為什麼 Speechable 外掛需要外部資源？
- 文本轉語音引擎需要機器學習運行時（ONNX）和過大的語音模型，無法捆綁到插件中。這些資源按需從CDN服務加載，類似Google字體或其他網絡服務的工作方式。這種基於服務的方法使得插件輕量化，同時提供高質量的AI語音。
外掛標籤
開發者團隊
📦 歷史版本下載
原文外掛簡介
Speechable transforms your WordPress content into natural-sounding audio using AI-powered text-to-speech technology. Speechable uses Piper TTS, an open-source neural text-to-speech engine that runs in your browser.
Third-Party Service
This plugin uses the following third-party services for text-to-speech functionality:
Piper TTS Web Library
Service Provider: Mintplex Labs
Library CDN: jsDelivr (https://cdn.jsdelivr.net/npm/@mintplex-labs/piper-tts-web)
ONNX Runtime CDN: Cloudflare (https://cdnjs.cloudflare.com/ajax/libs/onnxruntime-web)
WASM Phonemizer CDN: jsDelivr (https://cdn.jsdelivr.net/npm/@diffusionstudio/piper-wasm)
Voice Models: Downloaded from Hugging Face
Terms of Service: MIT License
Whisper (Word Timestamps)
Service Provider: Hugging Face Transformers.js
Library CDN: jsDelivr (https://cdn.jsdelivr.net/npm/@huggingface/transformers)
Model: Xenova/whisper-tiny.en (~75MB, downloaded on first use)
Purpose: Provides accurate word-level timestamps for synchronized highlighting
Terms of Service: Apache 2.0 License
What data is transmitted:
The TTS library and ONNX runtime are loaded from CDN when generating audio
Voice model files are downloaded from Hugging Face when first using a voice
Whisper model is downloaded for word timestamp extraction (first use only)
Your content text is processed locally in your browser and is NOT sent to any external server
Once downloaded, all resources are cached locally in your browser
Why external resources are required:
The text-to-speech engine requires machine learning runtime (ONNX) and voice models that are too large to bundle with the plugin. These are loaded on-demand from CDN services, similar to how Google Fonts or other web services work. This is a service-based approach that keeps the plugin lightweight while providing high-quality AI voices.
Key Features
AI-Powered Voices – Natural-sounding speech using Piper TTS neural network models
Accurate Word Highlighting – Whisper-powered word timestamps for precise synchronization
12 Languages – English, German, French, Spanish, Italian, Portuguese, Dutch, Polish, Russian, Chinese, Japanese, Korean
Customizable Player – Match your theme with custom colors and styling
Download Audio – Let visitors download audio files for offline listening
Voice Presets – Quick effects like Radio, Stadium, Cave, Chipmunk, Robot, and more
Free – No premium version, no limits, no hidden costs
How It Works
Install and activate the plugin
Go to Settings → Speechable to configure your preferences
Edit any post and click “Generate Audio” in the sidebar
The audio player automatically appears on your published posts
Voice Presets
Transform your audio with one-click presets:
Default – Natural voice
Radio – Broadcast quality
Stadium – Echo effect
Cave – Deep reverb
Chipmunk – High-pitched fun
Deep – Low bass voice
Robot – Mechanical tone
Telephone – Phone quality
Megaphone – Announcement style
Giant – Booming voice
Fairy – Magical high voice
Narrator – Audiobook style
Perfect For
Bloggers – Give readers an audio option
News Sites – Accessibility for all visitors
Educational Content – Help students learn
Accessibility – Support visually impaired users
Podcasters – Quick audio versions of written content
Privacy
Your content is processed locally in the browser. Text is not sent to external servers. Voice models are downloaded once and cached locally.
Usage
Getting Started
After activation, go to Settings → Speechable to configure your default options
Select your preferred language and voice
Choose which post types should have audio (posts, pages, etc.)
Customize the player colors to match your theme
Generating Audio for a Post
Method 1: Block Editor (Single Post)
Edit any post or page in the WordPress block editor
Look for the “Speechable” panel in the right sidebar (under Document settings)
Select your preferred voice and quality settings
Click “Generate Audio” and wait for processing
Once complete, you can preview the audio before publishing
Publish or update your post – the audio player will appear automatically
Method 2: Posts List (Quick Access)
Go to Posts → All Posts
Hover over any post to see the “Audio” button in the row actions
Click it to open the audio generation modal
Configure settings and click “Generate”
The audio will be saved to that post
Configuring Settings
Voice Settings
Language – Choose from 12 supported languages
Voice – Select a voice for the chosen language (preview available)
Quality – Low (faster), Medium (balanced), or High (best sync accuracy)
Whisper Model – Controls word timestamp accuracy (Tiny is fastest, Small is more accurate)
Voice Presets
Click any preset to quickly apply pitch and reverb effects:
Default, Warm, Bright, Radio, Narrator, Podcast, Deep, Soft, Room, Hall, Telephone, Vintage
Display Settings
Post Types – Enable audio for posts, pages, or custom post types
Player Position – Show player before or after content
Word Highlighting – Enable/disable word-by-word highlighting during playback
Auto-scroll – Automatically scroll to keep highlighted word visible
Player Appearance
Customize background, text, button, progress bar, and highlight colors
Adjust border radius for rounded or square corners
Live preview shows your changes instantly
Managing Audio
Regenerate – Click “Regenerate” in the editor panel to create new audio with different settings
Delete – Remove audio from a post using the “Delete” button
Download – Visitors can download the audio file using the player’s download button
Tips for Best Results
Use High quality for important content where word sync matters
Preview voices in Settings before generating to find the best match
Shorter posts generate faster – consider breaking very long content into parts
First generation takes longer as voice models are downloaded and cached
Word highlighting works best with standard theme content areas
Credits
Core Technologies
Piper TTS – rhasspy/piper by Mike Hansen
Fast, local neural text-to-speech system (MIT License)
Piper TTS Web – Mintplex-Labs/piper-tts-web v1.0.4
Browser-compatible Piper TTS implementation (MIT License)
OpenAI Whisper – openai/whisper
Robust speech recognition for word-level timestamps (MIT License)
Transformers.js – huggingface/transformers.js v3.4.1
Run Whisper models in the browser by Xenova (Apache 2.0 License)
ONNX Runtime Web – microsoft/onnxruntime
Machine learning inference engine (MIT License)
Piper WASM Phonemizer – diffusionstudio/piper-wasm
WebAssembly phonemizer for text processing (MIT License)
Voice Models
Piper Voices – diffusionstudio/piper-voices
High-quality neural voice models hosted on Hugging Face
Whisper Models – Xenova on Hugging Face
Optimized Whisper models for browser inference (whisper-tiny.en, whisper-small)
Infrastructure
jsDelivr CDN – Fast, reliable CDN for npm packages
Cloudflare CDN – ONNX Runtime delivery
Hugging Face – Model hosting and distribution
Design
Lucide Icons – lucide.dev (ISC License)
Beautiful, consistent icon set
Development
Developer: Tanish Mittal
Plugin Architecture: WordPress standards-compliant
Browser Compatibility: Modern ES6+ with Web Workers
