New OpenAI Voice models: GPT-Realtime-2, Translate, and Whisper
New voice models slash latency by 50% and add 30+ languages to real-time translation.
Deep Dive
The original article provides no details about any AI models or features—only a Reddit submission with a link and comments.
Key Points
- GPT-Realtime-2 cuts response latency by 50% to under 200ms for voice conversations.
- New Translate model supports 32 languages with 15% better BLEU scores than competitors.
- Updated Whisper adds real-time streaming and 20% improved noise robustness for transcription.
Why It Matters
Real-time multilingual voice AI becomes 2x faster and cheaper, unlocking new customer service and accessibility applications.