88Z AI, 13 November — iFlytek has launched SparkAudio 1.0, its first voice foundation model trained on fully domestic computing power. The model supports multi-language and dialect recognition, offers deep audio understanding capabilities, and achieves significant performance improvements and cost reductions through three technical innovations.
At a glance
- SparkAudio 1.0 supports 99 languages and 202 dialects, with an English recognition error rate of only 2.25%.
- Three technical innovations improve model performance by 20%, increase response speed by 15%, and reduce costs by 60%.
The update
iFlytek has officially launched SparkAudio 1.0 on its open platform, marking the first voice foundation model trained using fully domestic computing power. The model supports recognition for 99 languages and 202 dialects, achieving an English recognition error rate of just 2.25%. SparkAudio 1.0 features deep audio understanding capabilities, allowing it to perceive emotions, ambient sounds, and speaker attributes, while also supporting audio-based question answering. By incorporating three technical innovations—semi-supervised audio encoding, dynamic segmentation end-to-end modeling, and dual-paradigm collaborative decoding—the model improves overall performance by 20%, accelerates response speed by 15%, and reduces costs by 60%. This launch represents a further breakthrough in iFlytek's voice technology, offering developers a more efficient and cost-effective voice interaction solution.
Sources & editorial note
This article is an AI-assisted synthesis by 88Z AI based on the public sources below. Product claims and availability are attributed to their publishers. It does not represent independent testing or first-hand reporting.
- 讯飞开放平台Source published
Source dates above refer to the original announcements; the date under the headline is this article’s publication on 88Z AI.