Add support for speech-to-text models

Models like OpenAI’s Whisper provide speech to text capabilities. This is useful if you want to create text form of audio content like podcasts or generate subtitles for videos.

The php-ai-client package currently provides capabilities and options for Text To Speech models, but not for Speech To Text.

Currently available:

• CapabilityEnum::textToSpeechConversion()
  • https://github.com/WordPress/php-ai-client/blob/48cc7de403e00d3035ce2fcb88128dea5e283444/src/Providers/Models/Enums/CapabilityEnum.php#L16
• OptionEnum::outputMimeType()
  • https://github.com/WordPress/php-ai-client/blob/48cc7de403e00d3035ce2fcb88128dea5e283444/src/Providers/Models/Enums/OptionEnum.php#L31
• ModalityEnum::audio()
  • https://github.com/WordPress/php-ai-client/blob/48cc7de403e00d3035ce2fcb88128dea5e283444/src/Messages/Enums/ModalityEnum.php#L17

Support for Speech To Text

To support Speech To Text models, I’d propose something like this:

Add:

• CapabilityEnum::speechToTextConversion()
• OptionEnum::inputMimeType()

Usage:

	    $sttCapabilities = [
		    CapabilityEnum::speechToTextConversion(),
	    ];
	    $sttOptions = [
		    new SupportedOption(OptionEnum::inputModalities(), [[ModalityEnum::audio()]]),
		    new SupportedOption(OptionEnum::inputMimeType(), [
			    'audio/mpeg',
			    'audio/ogg',
			    'audio/wav',
			    'audio/flac',
			    'audio/aac',
		    ]),
		    new SupportedOption(OptionEnum::outputModalities(), [[ModalityEnum::text()]]),
                    new SupportedOption(OptionEnum::outputMimeType(), ['text/plain', 'application/json']),
		    new SupportedOption(OptionEnum::customOptions()),
	    ];

(Inspired by OpenAI TTS settings).

Start with src/Providers/Models/Enums/CapabilityEnum.php and OptionEnum.php, then compare the existing metadata definitions in src/ProviderImplementations/OpenAi/OpenAiModelMetadataDirectory.php. Add the proposed speech-to-text capability and input MIME type consistently with the existing enums and metadata; done means speech-to-text models can advertise the listed input and output modalities and formats.