Tsinghua university's iFlytek launches sparkaudio 1.0 voice foundation model API, supporting 99 languages and 202 dialects

iFlytek announced the release of its SparkAudio 1.0 voice foundation model API service on its open platform. This new model supports both text and voice input, with text output capabilities, enabling various applications such as speech transcription, multilingual translation, dialect recognition, environmental sound understanding, speaker separation, emotion analysis, and complex audio question answering. This advancement addresses limitations found in traditional cascaded audio processing methods, which often result in information loss—such as the omission of tone or background noise—and system fragmentation, leading to latency and errors. SparkAudio 1.0 directly interprets audio, understanding not just the spoken words but also the underlying semantics, emotion, and scene context. Compared to its predecessor, SparkAudio 1.0 offers significant upgrades, notably supporting 99 languages and 202 dialects for enhanced multilingual tasks. Furthermore, its deeper comprehension capabilities allow it to be applied across diverse sectors, including intelligent customer service, automotive cockpits, cross-language meetings, medical record analysis, and content moderation.