tts-1-hd

Overview. TTS-1 HD is OpenAI's high-definition text-to-speech synthesis model, designed to prioritize audio fidelity and natural articulation over generation speed. Released alongside the standard TTS-1 model, it transforms character strings into spoken audio across nine built-in voice profiles. Developers select this model when output audio quality takes precedence over sub-second streaming responsiveness.

Key strengths. The model focuses on clear pronunciation and reduced acoustic artifacts across complex spoken passages. It supports multiple audio encoding containers and raw formats—including MP3, Opus, AAC, FLAC, WAV, and PCM—and provides consistent voice styling across options such as alloy, ash, coral, echo, fable, onyx, nova, sage, and shimmer.

When to use it. Use TTS-1 HD for high-fidelity spoken content, automated voiceover production, audiobook narration, and interactive applications where acoustic polish is essential. For live conversational dialogue or latency-critical voice bots, route instead to lower-latency models or its designated successor, gpt-realtime-2.1-mini.

Limitations. The model enforces a hard request ceiling of 4,096 characters and cannot process inbound audio, images, or documents. As a dedicated audio generation endpoint, it does not support tool calling, structured outputs, or text completions. OpenAI has formally deprecated TTS-1 HD, with an end-of-life shutdown scheduled for January 6, 2027.

Sovereignty note. TTS-1 HD is a proprietary service operated by OpenAI in the United States, subjecting inference workloads to US legal jurisdictions, including the CLOUD Act. It does not qualify for Sovereign classification.