
Hugging Face has unveiled Real World VoiceEQ, a benchmark aimed at assessing the human quality of voice AI systems. This new evaluation tool goes beyond traditional metrics like word error rates, focusing instead on how well voice models can handle nuances such as tone, emotion, and speaker identity. The benchmark covers over 40 models and uses data from more than a million human ratings. By emphasizing real-world conversational dynamics, VoiceEQ seeks to advance voice AI towards more natural and reliable interactions.
Read original
© Hugging Face BlogNVIDIA's Nemotron 3 Embed models have set a new benchmark in retrieval quality, with the 8B model ranking #1 on the RTEB leaderboard. This collection of embedding models is designed for production-scale retrieval tasks, offering open weights and datasets for customization. The models support multilingual and code retrieval, and are optimized for high-throughput deployment with NVIDIA's NVFP4 technology. This release provides developers with powerful tools for efficient and accurate retrieval, enhancing capabilities in agentic retrieval and reducing operational costs.
© Hugging Face BlogDharmaOCR has demonstrated a significant advantage over newer OCR models like Mistral OCR4 and Unlimited-OCR in handling Brazilian Portuguese documents. This success is attributed to its domain-specific training, which focuses entirely on the linguistic nuances of Brazilian Portuguese. By employing a two-stage training process, including Direct Preference Optimization, DharmaOCR achieves higher extraction quality and stability. This specialization allows it to outperform more generalized models, highlighting the benefits of targeted training over broader multilingual approaches.