Smart Turn v3.2: Handling noisy environments and short responses

Smart Turn v3.2 is an open-source AI voice agent turn detection model that offers significant advancements in handling short utterances and background noise, achieving 40% better accuracy for short responses and improved robustness in noisy environments. The model, available on HuggingFace, comes with complete weights, datasets, and training code, making it accessible for developers to integrate as a drop-in replacement for v3.1. Enhancements include a new dataset for short utterances and a fix for a padding issue during training, alongside the inclusion of realistic background noise in the datasets to enhance performance in real-world scenarios. Users can implement this version with Pipecat's LocalSmartTurnAnalyzerV3, and further details along with benchmarks are available on the project's GitHub repository.