At this afternoon's Global 1024 Developer Festival, iFlytek officially unveiled the first voice cornerstone of the Spark family — the Spark voice foundation large model SparkAudio 1.0. It weaves machine translation, speech recognition, and semantic understanding into a single model, lifting fundamental voice capabilities up a notch overall.
In the official demo, SparkAudio 1.0 used a movie clip to show just how sharp its "ears" are: it not only precisely distinguished sound content, background, and the speaker's gender, but also made accurate judgments about the scene. According to iFlytek, it "performs excellently" in real-time speaker diarization, environment understanding, audio understanding, and emotion recognition, pushing AI one step further toward applications in the real physical world.
This new cornerstone has already been embedded in several iFlytek products. iFlytek Input Method 16.0, iFlytek Conference, iFlytek Office Notebook, and iFlytek Translator have all gone live, supporting free switching between Chinese and English and real-time speaker recognition — in other words, if you jump between two languages while speaking, with several people around speaking one after another, it can keep up and tell them apart.
Companion hardware was also presented. iFlytek released the "iFlytek AI Recording Ring," equipped with a 4+1 ring microphone array and using Wi-Fi 6 transmission, claiming an overall recognition accuracy of over 96%, and a 2-hour recording transcription can be completed in just 2 minutes.