OpenAI launches new voice intelligence features in its API

OpenAI said Thursday that its API volition present see a fig of caller dependable quality features designed to assistance developers make apps that tin talk, transcribe, and construe conversations with users.

The company’s new GPT‑Realtime‑2 is different dependable model, built to make a realistic vocal simulation that tin converse with users. However, dissimilar its predecessor (GPT-Realtime-1.5) this 1 is built with GPT‑5‑class reasoning that OpenAI says was created to woody with much analyzable requests from users.

The institution is besides launching GPT‑Realtime‑Translate which, conscionable arsenic it sounds, is designed to supply real-time translation services that “keep pace” with the user, conversationally. The diagnostic includes much than 70 input languages (that is, the languages that it tin comprehend) and 13 output languages (the languages it relays to the speaker).

Finally, the institution has besides launched a caller transcription capability, GPT-Realtime-Whisper, which gives users unrecorded speech-to-text capabilities that are captured arsenic interactions occur.

“Together, the models we are launching determination real-time audio from elemental call-and-response toward dependable interfaces that tin really bash work: listen, reason, translate, transcribe, and instrumentality enactment arsenic a speech unfolds,” the institution said.

Who volition these updates beryllium bully for? Companies that privation to grow lawsuit work capabilities are an evident target. However, OpenAI besides notes that its caller features volition assistance with a wide array of areas, including education, media, events, and creator platforms, among others.

As utile arsenic these tools look from an endeavor perspective, it besides seems plausible that they could beryllium misused. The institution said it has built guardrails to halt its caller features from being abused to make spam, fraud, oregon different forms of online abuse. Certain triggers person been embedded successful the strategy truthful that “conversations tin beryllium halted if they are detected arsenic violating our harmful contented guidelines,” OpenAI said.

All of the caller dependable models are included successful OpenAI’s Realtime API. Translate and Whisper are billed by the minute, portion GPT-Realtime-2 is billed by token consumption.