LS AI Provider for ElevenLabs
| 开发者 |
laurisaarni
whyisjake |
|---|---|
| 更新时间 | 2026年9月14日 20:14 |
| PHP版本: | 7.4 及以上 |
| WordPress版本: | 7.1 |
| 版权: | GPL-2.0-or-later |
| 版权网址: | 版权信息 |
详情介绍:
- Text-to-speech conversion with high-quality ElevenLabs voices
- Automatic voice selection -- a prompt works without configuring a voice ID first
- Long-form narration -- text beyond the model's per-request limit is narrated across several requests and returned as one audio file
- Sound effects generation from text descriptions
- Voice directory for discovering available voices (including cloned voices), cached per API key
- Dynamic model discovery from the ElevenLabs API
- Automatic provider registration
TEXT_TO_SPEECH_CONVERSION-- convert text to speech using any ElevenLabs TTS modelSPEECH_GENERATION-- generate sound effects from text prompts
- PHP 7.4 or higher
- The PHP AI Client SDK must be loadable. WordPress 7.0 and later bundle it in core; earlier WordPress needs it provided via Composer (it is an SDK, not a plugin)
- An ElevenLabs account and API key
安装:
- Ensure the PHP AI Client SDK is available (bundled in WordPress 7.0+)
- Upload the plugin files to
/wp-content/plugins/ls-ai-provider-for-elevenlabs/ - Activate the plugin through the 'Plugins' menu in WordPress
- Configure your ElevenLabs API key in Settings > Connectors (WordPress 7.0+), or via the
ELEVENLABS_API_KEYenvironment variable or constant
常见问题:
How do I get an ElevenLabs API key?
Visit https://elevenlabs.io/app/settings/api-keys to create an account and generate an API key. Using this plugin requires an ElevenLabs account, and ElevenLabs bills you for the requests it makes on your behalf.
Does this plugin work without the PHP AI Client SDK?
No. This plugin contains only the ElevenLabs-specific implementation of the AI Client's provider interfaces, so the SDK has to be loadable for it to do anything. WordPress 7.0 and later bundle the SDK in core. On earlier WordPress, the SDK has to be provided by your project (for example through Composer); it is a library, not a plugin. When the SDK is missing, this plugin registers nothing and stays inert instead of erroring.
How do I specify which voice to use?
Set the outputSpeechVoice option in your ModelConfig to the voice ID. You can discover available voices using the VoiceDirectory class or the ElevenLabs voice library.
What happens if I don't specify a voice?
The provider resolves a default: the ELEVENLABS_DEFAULT_VOICE_ID environment variable or constant, the ai_provider_for_elevenlabs_default_voice_id option, a voice discovered from your own ElevenLabs account (preferring premade voices; requires the Voices permission on the API key), and finally the premade voice "George" (JBFqnCBsd6RMkjVDRZzb), which is available on every account. The result passes through the ai_provider_for_elevenlabs_default_voice_id filter. An explicitly configured outputSpeechVoice always takes precedence.
Can it narrate a whole post?
Text longer than the model's per-request character limit is split on paragraph and sentence boundaries, narrated in several requests, and returned as one audio file. Note that narration is slow (roughly 90 to 95 characters per second) and runs inside one PHP request, so very long text can exceed the PHP execution time limit. Each chunk is billed as its own API request. See the README for details and for the public per-chunk methods a background-processing plugin can build on.
Why does the AI plugin say there is no valid AI Connector?
The WordPress.org AI plugin treats a connector as valid only when it can generate text, and ElevenLabs generates speech and sound. The warning is expected and does not affect this provider; add a text-generation connector alongside it to satisfy the AI plugin's own features.
What audio formats are supported?
The default output format is MP3 (mp3_44100_128). Other supported formats include PCM, ulaw, Opus, and AAC at various sample rates and bitrates. When long text has to be split across requests, only formats whose audio can be joined are allowed (MP3, PCM, ulaw, alaw).
Does the plugin send anything anywhere on its own?
No. Nothing leaves your site until code on your site asks the AI Client for speech, sound, a model list, or a voice list. There is no telemetry, no analytics, and no phone-home. The "External services" section below documents every request the plugin can make.
更新日志:
- Rename the plugin to "LS AI Provider for ElevenLabs", with the permalink
ls-ai-provider-for-elevenlabs. The WordPress.org plugin directory requires a plugin name to begin with a distinctive identifier, and "AI Provider" on its own is a generic description; "LS" is the author's initials - Rename the bootstrap to
ls-ai-provider-for-elevenlabs.phpand the text domain tols-ai-provider-for-elevenlabs, both of which have to match the permalink - Keep the filter, option and transient names (
ai_provider_for_elevenlabs_*) and the Composer package name (saarnilauri/ai-provider-for-elevenlabs) unchanged, so existing integrations and Composer installs are unaffected
- Verify the API key against ElevenLabs instead of only checking that one was entered, so Settings > Connectors reports whether the key actually works. A key rejected by ElevenLabs now shows as not connected, where any non-empty value previously showed as connected
- Treat a key scoped without the Models permission as working rather than broken. Such a key still drives text-to-speech, and the plugin already falls back to its built-in model list, so the connection check accepts it. The verification result is cached for 15 minutes, under a name derived from a hash of the key rather than the key itself
- Stop reading the core Connectors option (
connectors_ai_elevenlabs_api_key) directly. WordPress passes the key to the AI Client itself, and the plugin already re-wraps it for thexi-api-keyheader ElevenLabs requires, so the direct read was redundant. This resolves a WordPress.org review finding about plugins handling AI Client credentials
- First release from the WordPress.org plugin directory
- Rename the plugin bootstrap from
plugin.phptoai-provider-for-elevenlabs.php, so the plugin headers live in the file matching the plugin slug as the WordPress.org directory expects. Sites that installed an earlier build from the GitHub ZIP have to reactivate the plugin once after updating - Document every ElevenLabs endpoint the plugin can call, what is sent to it and when, in a new "External services" section of the readme
- Correct the readme's claim that a separate "PHP AI Client plugin" has to be installed: the AI Client is an SDK, bundled in WordPress 7.0 and later
- Redraw the provider logo shown on the Connectors screen from the official ElevenLabs symbol, at the proportions and clear space their brand guidelines specify
- Add the WordPress.org directory banner and icon, built from the official ElevenLabs artwork in their monochrome palette, a WordPress Playground blueprint for the directory preview, and a Plugin Check run in continuous integration
- Long-form narration: text beyond the model's per-request character limit is split on paragraph and sentence boundaries, narrated across several requests carrying neighbouring text for prosody, and returned as one audio file (contributed by Jake Spurlock)
- Automatic voice selection: when no
outputSpeechVoiceand no explicit default are configured, a voice is discovered from the account's own voices, preferring premade ones, before falling back to "George" (contributed by Jake Spurlock) - Voice directory: migrate from the deprecated
/v1/voicesendpoint to/v2/voiceswith full pagination, and cache the voice list per API key for 15 minutes (contributed by Jake Spurlock) - Honour all
customOptionsin both the text-to-speech and sound generation models (e.g.language_code,seed,speed,apply_text_normalization); previously most keys were silently dropped (contributed by Jake Spurlock) - Read each model's real per-request character limit from the live
/modelsresponse, seeded from measured values, and addeleven_v3to the fallback model list (contributed by Jake Spurlock) - Fix the voice directory staying unusable for the rest of the request when built before credentials were available (contributed by Jake Spurlock)
- Show the provider in the connector UI as "ElevenLabs" with a description and the official logo (contributed by Jake Spurlock)
- Exclude local credentials (
.env,.wp-env.override.json) and dev artifacts from the release ZIP, verified by a CI canary check (contributed by Jake Spurlock) - Add continuous integration (unit tests on PHP 7.4-8.4, phpcs, PHPStan, packaging leak check), a local
wp-envenvironment, the GPL-2.0 license text, and fix test-suite autoloading on case-sensitive filesystems
- Declare inline
outputFileTypesupport for text-to-speech and sound generation models, so support checks likeisSupportedForTextToSpeechConversion()pass when callers request inline output (fixes compatibility with the WordPress AI plugin's Text to Speech experiment) - Fall back to the premade "George" voice when no
outputSpeechVoiceis configured, instead of throwing an exception; the default can be overridden via theELEVENLABS_DEFAULT_VOICE_IDenvironment variable or constant, theai_provider_for_elevenlabs_default_voice_idoption, or theai_provider_for_elevenlabs_default_voice_idfilter
- WordPress 7.0 compatibility: read the API key from the core Connectors option (
connectors_ai_elevenlabs_api_key, Settings > Connectors) - WordPress 7.0 compatibility: move the xi-api-key authentication restore hook to init priority 21, after core's Connectors credential pass (init 20) which overwrites provider auth with generic Bearer authentication
- Fix API key validation on the WordPress 7.0 Connectors page: the availability check now also considers the registry request authentication and the Connectors/legacy credential options, instead of only the ELEVENLABS_API_KEY environment variable or constant
- Fix issue if API key not yet set in options table
- Fix API key registration, if the API key is not set in constant, but in the connectors page in WordPress admin
- Add a workaround to restore ElevenLabs authentication when the AI client initiation overwrites it with default auth.
- Initial release
- Text-to-speech conversion with ElevenLabs TTS models
- Sound effects generation from text prompts
- Voice directory for listing and discovering available voices
- Dynamic model discovery from the ElevenLabs API
- Custom voice settings support (stability, similarity_boost, style, use_speaker_boost)
- Multiple output format support (MP3, PCM, Opus, AAC, ulaw)