#OpenAI also released other new features, such as #WebRTC support [4] and increased maximum session length from 15 to 30 minutes [5].

[4] https://platform.openai.com/docs/guides/realtime-webrtc

[5] https://platform.openai.com/docs/guides/realtime-model-capabilities#session-lifecycle-events

#GenAI #VoiceBots #Chatbots #AI #LLMs #Agents #RealtimeAPI

💸 #OpenAI is going after one of the major pain points of its audio-native models: price. The newest audio model, `gpt-4o-realtime-preview-2024-12-17`, will cost 60% less than its predecessor. #gpt-4o-mini also becomes available through Realtime API, at 10x cheaper cost per token than the old gpt-4o-realtime (10$/M token input, 20$/M tokens output) [3].

[3] https://openai.com/api/pricing/

#GenAI #VoiceBots #Chatbots #AI #LLMs #Agents #RealtimeAPI

🚦 #LiveKit released a transformers-based, semantic End-of-Turn detector, #opensource on #HuggingFace[1]! This model complements voice activity detectors (#VAD) by predicting whether the user's sentence is complete. This helps reduce false starts up to 85% according to their own testing, and is text-based, with a very low latency (~50ms). Find all the details in their post [2].

[1] https://huggingface.co/livekit/turn-detector

[2] https://blog.livekit.io/using-a-transformer-to-improve-end-of-turn-detection

#GenAI #VoiceBots #Chatbots #AI #LLMs #Agents #RealtimeAPI