ai@7.0.150
Vercel AI SDK 7.0.150 introduces experimental streamSpeech functionality to stream speech audio from text streams or collections using compatible Experimental StreamingSpeechModelV4 models.
原始内容为英文；当前页面提供中文导航与来源说明，具体事实请以原文为准。
为什么值得关注
The patch release ai@7.0.150 adds an experimental streamSpeech feature. It allows streaming speech audio from strings, arrays, iterables, or streams of text deltas and parts (with optional voice and providerOptions settings) via models implementing the experimental Experimental StreamingSpeechModelV4 interface. The method provides single-consumer audioStream and fullStream, in addition to promises for usage, warnings, responses, and providerMetadata. Provider error parts do not fatally terminate the process and are routed to an onError handler (defaulting to console.error), whereas user aborts or cancels reject result promises immediately. The update also introduces Experimental SpeechTextPart, Experimental SpeechStreamPart, Experimental SpeechStreamUsage, and Experimental StreamSpeechResult types, an optional cause property to NoSpeechGeneratedError, and Experimental MockStreamingSpeechModelV4 in ai/test. Dependency bumps include @ai-sdk/gateway@4.0.123, @ai-sdk/provider-utils@5.0.67, and @ai-sdk/provider@4.0.33.
可执行摘要
Vercel AI SDK v7.0.150 adds experimental streamSpeech for streaming speech audio from strings, iterables, or streams of text parts. It outputs single-consumer audioStream and fullStream along with usage, warnings, responses, and providerMetadata promises, supported by new experimental types and a mock testing model.
- Agent 实用度
- 77/100
- 可信度
- 96%
- 机器格式
- JSON + Markdown
开发者应核对什么
- Upgrade to ai@7.0.150 if streaming text-to-speech functionality is needed.
- Utilize Experimental MockStreamingSpeechModelV4 from ai/test when writing tests for speech streaming pipelines.
- Handle non-fatal streaming errors via the onError callback option in streamSpeech.