iFLYTEK launches SparkAudio 1.0 speech model API supporting 99 languages

iFLYTEK has launched API access for its SparkAudio 1.0 speech foundation model on its open platform, and developers can plug in directly. The model takes voice and text as input and outputs text, covering transcription, translation, dialect recognition, sound scene understanding and speaker separation.

In the past, audio was mostly processed step by step: first turning speech into text, then letting a model read it. This loses tone, emotion and background sounds, and chained modules often stack up errors and lag. SparkAudio 1.0 understands voice directly, picking up speech, sounds and music at once, along with meaning, emotion and scene.

Compared with the last version, the new model supports 99 languages and 202 dialects, and works for smart customer service, car cockpits and cross-language meetings. It is trained on Chinese domestic computing power, offering a self-controlled voice AI base for industries.