iFlytek SparkAudio 1.0 Launches API with 99-Language Support

On November 12, 2026, IT Home reported that iFLYTEK had announced the release of its Spark voice foundation model, SparkAudio 1.0, as an API service on the iFLYTEK Open Platform, where developers can integrate and call it directly.

According to the company, SparkAudio 1.0 accepts both text and speech as input and produces text as output, covering tasks such as speech-to-text transcription, multilingual translation, multi-dialect recognition, environmental sound understanding, speaker separation, sentiment analysis, and complex audio question-and-answer scenarios. iFLYTEK explained that audio processing has traditionally relied on a cascaded approach, in which speech is converted to text first and the text is then handed to an LLM for understanding. The company argued that this setup suffers from two drawbacks: text captures only what was said, so tone, emotion, and background sound are lost, and the pipeline links separate capabilities such as transcription, voiceprint recognition, and speech translation into independent modules, which creates breakpoints that accumulate errors and introduce latency and stuttering.

iFLYTEK stated that SparkAudio 1.0 moves beyond speech-to-text by directly understanding audio, taking in human voices, environmental sounds, and music in one pass while interpreting the semantics, emotion, and scene carried in sound, and said the model brings fresh upgrades across multiple areas compared with the previous generation. The report's headline states that the model supports 99 languages and 202 dialects. At this point the available information extends only to the API launch; no further developments have been reported.