~/ARTIFICIAL I/iflytek-launches-sparkaudio-1-0-native-speech-model-api

iFlytek Launches SparkAudio 1.0 Native Speech Model API

iFlytek has launched API services for its SparkAudio 1.0 native speech foundational model on the iFlytek Open Platform, supporting text and speech inputs alongside text output. The model provides direct audio-to-meaning processing capabilities across 99 languages and 202 dialects. This release marks a shift from traditional cascaded speech systems to native audio foundational models that can directly process tone, emotion, and background acoustics without information loss. It offers developers an autonomous and controllable speech intelligence foundation trained entirely on domestic computing power. SparkAudio 1.0 supports tasks such as speech transcription, multi-language translation, dialect recognition, environmental sound understanding, speaker diarization, sentiment analysis, and audio Q&A. It is trained on domestic computing power and is now accessible via official API endpoints.

## BACKGROUND

Traditional speech processing systems typically use a cascaded approach that first transcribes speech into text before sending it to a language model, which discards crucial non-verbal cues like tone and background noise. In contrast, native speech foundational models process audio signals directly to capture semantics, emotions, and environmental context simultaneously.

## REFERENCES

## KEYWORDS

#Artificial Intelligence#Audio Model#Speech Recognition#Multimodal AI#iFlytek

$ subscribe --daily