ai ai@7.0.150

This release adds an experimental way to turn text into speech while the text is still arriving. The new experimental_streamSpeech function in Vercel's AI SDK accepts a plain string, or an array, iterable or stream of text deltas and text parts, and each text part can carry its own voice and provider options. It only works with speech models that implement a new experimental streaming extension, Experimental_StreamingSpeechModelV4, and the release notes do not name any provider that implements it yet.

The function returns the audio as single-consumer streams (audioStream and fullStream), plus promises for usage, warnings, responses and provider metadata. Error parts reported by the provider are non-fatal; they go to an onError callback that defaults to console.error. A user abort or a cancel rejects the result promises immediately. The release also exports the related types, adds an optional cause to NoSpeechGeneratedError, and adds a mock streaming speech model, Experimental_MockStreamingSpeechModelV4, to ai/test for writing tests against the feature.