vercel/ai

Description

Add an official @ai-sdk/gradium provider package to the AI SDK.

Gradium provides low-latency Text-to-Speech and Speech-to-Text APIs for voice applications. The AI SDK already supports voice providers such as ElevenLabs and Deepgram, and adding Gradium would let developers use Gradium speech generation and transcription through the same AI SDK speech/transcription interfaces.

Proposed Scope

• Add @ai-sdk/gradium provider package.
• Support ProviderV4.
• Support speech generation via gradium.speech() / gradium.speechModel().
• Support transcription via gradium.transcription() / gradium.transcriptionModel().
• Add Gradium-specific provider options for voice IDs, output/input formats, json_config, rewrite rules, pronunciation dictionaries, and latency-related STT settings.
• Add helper APIs for voices, pronunciation dictionaries, and credits.
• Add tests and provider documentation.

Notes

Gradium also supports realtime WebSocket APIs with semantic VAD, adaptive delay, and flush-driven turn-taking. This provider focuses on the AI SDK request/response speech and transcription surfaces, while documenting when developers should use Gradium’s realtime APIs directly.

AI SDK Version

No response

Code of Conduct

• I agree to follow this project's Code of Conduct