Speech is Apple's framework, a library that is part of the operating system, for speech recognition: it turns spoken words into written text. It is available on iOS and iPadOS 10 and later, macOS 10.15 and later, Mac Catalyst and visionOS, and it gives an app the same kind of recognition that keyboard dictation uses, but without the keyboard, for example to understand spoken commands.
An app feeds it live audio from the microphone or audio from a file. In the current API, a SpeechAnalyzer object manages an analysis session and passes time-coded audio to modules: SpeechTranscriber transcribes normal conversation, DictationTranscriber works like system dictation and supports older devices, and SpeechDetector performs voice activity detection (finding the parts of the audio that contain speech). Results include the transcription, alternative interpretations and confidence levels. The AssetInventory class downloads the language-specific model assets the device needs, and custom language models can add a specialized vocabulary.
The framework has no separate price; distributing apps requires the paid Apple Developer Program.