Hands-on test reveals both strengths and gaps in SparkAudio 1.0
智东西 tested iFlytek's SparkAudio 1.0 with mixed audio and reported that it combined mechanical sounds with a child's description to infer that the surrounding environment could be a train or subway. The same test found that the model omitted a short warning sound and did not synthesize its conclusions across the two audio clips, limiting the demonstrated completeness of its scene analysis.
The hands-on findings provide concrete evidence of SparkAudio 1.0's mixed-sound reasoning capabilities while also documenting two limitations in the audio analysis demonstrated by 智东西.

智东西 作者 | 程茜 编辑 | 漠影
如果让你扮演一名刑侦助手,你能从下面数种声源混叠的音频里分析出受害人的所在位置吗?
讯飞最新发布的语音基座大模型SparkAudio 1.0,在几秒内就给出了答案:被拐孩子周围是大型机械设备发出的规律性噪音,可能在火车或地铁上。
模型给出了完整的分析过程,其没有根据孩子提到的“歌声”就默认是人或乐器发出的声音,而是通过嗡鸣声、机械设备的规律性噪音等数种声源,以及孩子提到的“在车上”去分析可能是什么场景发出的声音。不过,我们在音频中还能听到短促的警示音,模型并没有对此进行分析。