The body is code-first and lean; prose is limited to genuinely non-obvious notes (e.g. "Log write failures rather than throwing from the WebSocket callback thread") and does not re-explain concepts Claude already knows, so every token earns its place.

Provides fully executable, import-complete Java for REST, WebSocket, and async paths plus a verified parameter list, making the examples copy-paste ready rather than pseudocode.

Each synthesis flow is an unambiguous sequence (connect → sendText → sendFlush → sendClose with handler registration) with error-logging feedback; there are no destructive or batch operations that would demand additional validation checkpoints.

No bundle files exist, and the body is organized into clearly signaled one-level references — in-repo source paths and external OpenAPI/AsyncAPI/doc URLs — with no nested multi-hop references, matching the score-3 anchor for clear overview plus well-signaled navigation.

Names multiple concrete actions: "one-shot REST via `client.speak().v1().audio().generate(...)`" and "streaming synthesis via `client.speak().v1().v1WebSocket()`", matching the score-3 anchor listing several specific actions.

Explicitly answers both what (Deepgram TTS v1 audio synthesis via REST and WebSocket) and when ("Use when writing or reviewing Java code in this repo that calls Deepgram Text-to-Speech v1"), satisfying the score-3 anchor for both with explicit triggers.

Lists natural terms a user would say — "tts", "text to speech", "speak", "aura", "streaming tts", and "speak websocket" — giving good coverage of common variations rather than jargon.

Carves a clear Java/Deepgram-TTS-v1 niche and actively reduces conflict by redirecting full-duplex assistants to `deepgram-java-voice-agent`, so it is unlikely to trigger for the wrong skill.