← All changes | includes/AI/ModelRegistry.php +38 -0 4.9.0 → 4.9.4 View file →

		@@ -120,8 +120,46 @@
120	120	) );
121	121	}
122	122
123	123	/**
	124	+ * Speech-to-text model per platform, for audio and video attachments.
	125	+ *
	126	+ * Kept apart from the chat catalogue above because it is not a user choice:
	127	+ * the Settings model picker selects the model that *writes the doc*, while
	128	+ * transcription is a fixed preprocessing step. A platform absent from this
	129	+ * map cannot transcribe, which is what Core\WriteWithAI::platform_supports()
	130	+ * reads — so adding a key here is all it takes to light up a new provider.
	131	+ *
	132	+ * @since 4.9.4
	133	+ *
	134	+ * @return array<string,string>
	135	+ */
	136	+ public static function transcription_models() {
	137	+ return apply_filters( 'betterdocs_ai_transcription_models', array(
	138	+ // Cheaper and faster than whisper-1 at comparable accuracy, and it
	139	+ // accepts the same container list — including video, whose audio
	140	+ // track the endpoint extracts server-side.
	141	+ 'openai' => 'gpt-4o-mini-transcribe',
	142	+ // Gemini has no speech endpoint; the media rides inline in an
	143	+ // ordinary generateContent call, so this is a normal chat model.
	144	+ 'gemini' => 'gemini-flash-latest',
	145	+ ) );
	146	+ }
	147	+
	148	+ /**
	149	+ * The transcription model for a platform, or '' when it cannot transcribe.
	150	+ *
	151	+ * @since 4.9.4
	152	+ *
	153	+ * @param string $platform
	154	+ * @return string
	155	+ */
	156	+ public static function transcription_model( $platform ) {
	157	+ $models = self::transcription_models();
	158	+ return isset( $models[ $platform ] ) ? (string) $models[ $platform ] : '';
	159	+ }
	160	+
	161	+ /**
124	162	* Models for one platform (model_id => label). Empty array if unknown.
125	163	*
126	164	* @param string $platform
127	165	* @return array<string,string>