iFlytek releases Spark-ASR-2.0 speech recognition model
科大讯飞发布全新语音识别大模型 Spark-ASR-2.0,明日上线讯飞输入法
The model shifts speech recognition from verbatim transcription toward fluent, context-aware text generation, which changes how transcription-heavy workflows like meeting notes and dictation are built. Developers can call it through the iFlytek Open Platform API from 24 Sep, and its stated strength in dialects and noisy scenes is a concrete comparison point against other ASR offerings serving Chinese-language users.
On 23 Sep, iFlytek (科大讯飞) released Spark-ASR-2.0, its latest speech recognition large model. The model combines non-autoregressive and LLM-enhanced autoregressive decoding, joint Chinese-English text and acoustic enhancement, and dynamic context injection. It targets mixed Chinese-English speech, dialects, technical terms, and hard acoustic conditions such as high noise, low volume, fast speech and children's voices, with WER notably reduced versus Spark-ASR-1.0 and leading industry results. Inference cost rose only 10% over Spark-ASR-1.0. It will roll out in iFlytek Input Method starting 24 Sep and be offered via the iFlytek Open Platform API, later reaching AI glasses, smart office notebooks and iFlytek Tingjian.
科大讯飞发布全新语音识别大模型 Spark-ASR-2.0,明日上线讯飞输入法
IT之家 9 月 23 日消息,今日,科大讯飞正式发布最新一代语音识别大模型 Spark-ASR-2.0。 据官方介绍,Spark-ASR-2.0 通过非自回归与 LLM 增强自回归协同、中英文混合文本与声学联合增强、动态上下文注入等关键技术创新,整体的语音识别效果大幅提升,尤其在通用识别(中英文混合、方言、专业术语)、复杂声学场景识别(高噪、小音量、快语速、儿童)、上下文识别和文本流畅规范性等方面改善明显。在效果提升的同时,Spark-ASR-2.0 的整体推理成本相对 Spark-ASR-1.0 仅增加了 10%。 官方称,“如果说过去的语音识别追求的是一字不差地还原所说内容,那么 Spark-ASR-2.0 则更进一步,让识别结果‘流畅成文’—— 在提升识别准确率的同时,结合上下文逻辑与语境理解,精简冗余表达、精修各类细节,让文字更连贯、语义更清晰。” 相较于 Spark-ASR-1.0,Spark-ASR-2.0 在语音识别各核心场景上的效果有明显进步,尤其在中英文混合、方言、高噪声、文本流畅规范性等维度上的效果实现了大幅提升,WER(词错误率)明显降低。 Spark-ASR-2.0 与当前业界最好水平相比,在方言、高噪和小音量上拥有显著优势,也再次证明了在复杂声学场景上语音识别的强大实力,主要任务效果均好于业界最优水平。 测试集中 ASR 任务使用 WER / CER 作为评价指标,数值越低表示效果越好 * 测试集来自于语音真实任务请求数据,来源分布包括讯飞星火 APP、讯飞听见 APP、讯飞翻译机、讯飞输入法、讯飞开放平台语音 API 真实开发者场景等进行滚动更新。 据IT之家了解,Spark-ASR-2.0 将从明天开始逐步上线讯飞输入法,并且通过讯飞开放平台提供 API 服务。此外,Spark-ASR-2.0 还将陆续在讯飞 AI 眼镜、讯飞智能办公本、讯飞听见等更多产品业务上落地应用。