人民财讯8月22日电，8月22日，记者获悉，阿里通义发布新一代端到端的语音识别大模型Fun-ASR,该 模型增强了上下文感知和高精度语音转写能力，在家装、保险等多个行业场景的语音识别准确率均提升 了15%以上。目前，Fun-ASR已应用于会议字幕与同传、智能纪要、语音助手等场景，未来该模型将进 一步在阿里云百炼上线。 ...

Core Insights - The collaboration between DingTalk and Tongyi Laboratory has led to the launch of a new voice recognition model, Fun-ASR, which can understand industry-specific jargon across ten sectors, including home decoration and animal husbandry [1][2] - Fun-ASR has been integrated into various DingTalk functionalities such as meeting subtitles, simultaneous interpretation, smart minutes, and voice assistants [1] Technical Highlights - Fun-ASR enhances the recognition capability of industry-specific vocabulary, trained on over one hundred million hours of audio data, and co-created with real scenarios from DingTalk's multi-industry clients [1] - The model features improved contextual awareness and understanding, utilizing existing information within DingTalk, such as contact lists and schedules, to optimize inference and provide reliable transcription results [1] - Fun-ASR supports customized voice recognition model training for enterprises with advanced needs, allowing for algorithm optimization based on real scenario voice data provided by the companies [1] Future Plans - The voice team leader from Tongyi Laboratory expressed excitement about the partnership with DingTalk, aiming to expand the data and model scale of Fun-ASR to enhance the replicability of voice intelligence solutions for enterprise clients [2] - DingTalk's CTO highlighted the rapid development of Fun-ASR within three months of collaboration, achieving recognition from leading clients and marking a significant breakthrough towards industry leadership [2]