OpenAI: GPT Audio Mini

Converts text input to high-quality audio output using an upgraded neural decoder architecture that generates natural prosody, intonation, and voice characteristics. The model maintains consistent voice identity across multiple utterances by preserving speaker embeddings throughout the decoding process, enabling seamless multi-turn audio generation without voice drift or tonal inconsistency.

Provides access to a curated set of pre-trained voice profiles that can be selected via API parameter to generate audio with distinct speaker characteristics, accents, and tonal qualities. The model routes text input through voice-specific decoder pathways that apply learned speaker embeddings and acoustic characteristics, enabling developers to select appropriate voices for different use cases without managing separate models.

A lightweight variant of the full GPT Audio model that achieves lower per-request costs ($0.60 per million input tokens) through architectural optimizations including reduced model size, simplified decoder pathways, and efficient inference scheduling. The model maintains quality through selective parameter reduction while preserving the upgraded decoder for natural prosody, enabling cost-conscious deployments at scale without proportional quality degradation.

Supports chunked audio generation and streaming delivery via HTTP streaming responses, enabling clients to begin audio playback before the entire synthesis completes. The model generates audio in sequential chunks aligned to sentence or phrase boundaries, allowing progressive buffering and playback without waiting for full synthesis completion, reducing perceived latency in interactive applications.

Exposes text-to-speech functionality through a RESTful HTTP API with standardized JSON request format and audio file response, enabling integration into any application stack via standard HTTP clients. The API abstracts underlying model complexity through parameter-based configuration (voice selection, output format, speed), allowing developers to integrate audio generation without managing model infrastructure or dependencies.