Xiaomi releases industrial target speaker ASR model
小米开源工业级目标说话人语音识别大模型 Xiaomi-CocktailASR-1
For developers working on speech recognition systems, this model offers a high-performance solution for the challenging 'cocktail party' problem, with superior results compared to existing methods.
Xiaomi has released and open-sourced its industrial target speaker ASR model, Xiaomi-CocktailASR-1.
The model uses a target speaker's reference audio as a voiceprint prompt to extract and transcribe their speech in complex multi-speaker environments.
It achieves state-of-the-art results on multiple mainstream multi-speaker test sets, significantly outperforming existing solutions.
Its single-speaker recognition performance is on par with standard ASR models, and it can reject non-target speaker interference, outputting empty text when the target is not present.
小米开源工业级目标说话人语音识别大模型 Xiaomi-CocktailASR-1
IT之家 9 月 11 日消息，小米技术今日发文宣布，小米正式发布并开源工业级目标说话人语音识别大模型 Xiaomi-CocktailASR-1。 官方表示，Xiaomi-CocktailASR-1 旨在解决“鸡尾酒会”难题 （IT之家注：当前语音识别技术已经可以以较高精度识别一个人所讲的话，但是当说话的人数为两人或者多人时，语音识别率就会极大地降低，这一难题被称为鸡尾酒会问题）。 该模型采用端到端 LLM 架构，以目标说话人的一段参考音频作为声纹提示，可在多人同时说话的复杂环境中，精准提取并仅转录目标用户的语音。 在多个主流多说话人测试集上均达到 SOTA ，大幅领先现有方案。 同时， 其单人识别性能比肩标准 ASR 模型 ，可无缝兼顾单人说话场景；具备拒识非目标说话人干扰语音的能力，目标不在场时输出空文本，有效避免误触发；此外还支持思维链推理模式，为识别结果提供可解释推理过程。 IT之家附相关链接： GitHub： https://github.com/xiaomi-research/xiaomi-cocktailasr-1 HuggingFace： https://huggingface.co/Ease3/Xiaomi-CocktailASR-1
This report is categorized under AITOP's "Models" desk, focusing on China's AI industry and domestic foundation model ecosystem. Within our cross-regional analytical framework, it directly maps to AI developer toolchains, coding agents, and engineering workflows and US-China foundation model benchmarks and commercialization comparison. Key entities and technical topics under continuous tracking include Xiaomi-CocktailASR-1, target speaker ASR, cocktail party problem, open-source models.
Source attribution: curated from IT之家.
