Add voice resolution and defaulting to the Text to Speech experiment

[saarnilauri]

What?

Follows up on the ElevenLabs discussion in WordPress#888. Adds a Voice_Resolver to the Text to Speech experiment that dynamically resolves the voices a provider supports and defaults to the first one when no voice is configured.

Why?

Some providers (e.g. ElevenLabs) require a voice ID for text to speech. The experiment currently only sends a voice when the free-text Voice setting is filled in, so those providers fail out of the box with a raw SDK exception.

How?

• New Voice_Resolver reads the resolved TTS model's metadata for supported values declared on the outputSpeechVoice option (mirroring Speech_Generator's provider/model resolution order, cached in a transient for an hour). This establishes a generic contract: providers can declare their voice IDs as supportedValues on SupportedOption( OptionEnum::outputSpeechVoice() ).
• When voices are declared, the Voice setting renders as a select (existing elements/DataForm pattern, no JS changes) with a "Provider default (first available voice)" option, and Speech_Generator::generate_chunk() fills an empty voice with the first declared one — covering both the cron path and the ai/speech-generation ability.
• When no voices are declared (all providers today), the field stays free-text and behavior is unchanged, except:
  • a new wpai_tts_default_voice filter lets site or provider plugins supply a default voice, and
  • the raw "outputSpeechVoice is required" exception is mapped to an actionable error pointing at the Voice setting.

Note: the ElevenLabs provider 0.3.0 has been released with the complementary fixes — it now declares inline outputFileType support and falls back to a premade default voice when none is configured. The two fixes work together: an explicit voice from this plugin always overrides the provider default.

Use of AI Tools

AI assistance: Yes
Tool(s): Claude Code
Model(s): Fable 5
Used for: Planning and implementation. Review and testing done by me

Testing Instructions

1. Connect a provider that requires a voice (e.g. the ElevenLabs provider, version 0.3.0 or later) and enable the Text to Speech experiment
2. With the Voice setting empty, generate audio for a post — with ElevenLabs 0.3.0's default-voice fallback it succeeds; with a voice-requiring provider that has no default, the editor now shows the actionable "requires a voice" error instead of a raw exception
3. Add add_filter( 'wpai_tts_default_voice', fn () => '21m00Tcm4TlvDq8ikWAM' ); — generation succeeds via the cron path and via POST /wp-json/wp-abilities/v1/abilities/ai/speech-generation/run
4. Enter a voice ID in the Voice setting — it takes precedence
5. Register a test provider whose TTS model declares SupportedOption( OptionEnum::outputSpeechVoice(), array( 'v1', 'v2' ) ) — the Voice setting renders as a select and leaving it on "Provider default" uses v1
6. With OpenAI connected, verify no regression (empty voice → provider default path)

Changelog Entry

Added - Voice selection for the Text to Speech experiment: voices declared by the provider render as a dropdown, the first declared voice is used by default, and a wpai_tts_default_voice filter allows supplying a default for providers that require one.

🤖 PR message Partly generated with Claude Code

[github-actions]

The following accounts have interacted with this PR and/or linked issues. I will continue to update these lists as activity occurs. You can also manually ask me to refresh this list by adding the props-bot label.

If you're merging code through a pull request on GitHub, copy and paste the following into the bottom of the merge commit message.

Co-authored-by: saarnilauri <laurisaarni@git.wordpress.org>
Co-authored-by: dkotter <dkotter@git.wordpress.org>

To understand the WordPress project's expectations around crediting contributors, please review the Contributor Attribution page in the Core Handbook.