Newshot iFlytek Spark Speech Foundation Large Model SparkAudio 1.0 Launches API Service, Supporting 99 Languages and 202 Dialects
IT Home, November 12: iFlytek announced today that the iFlytek Spark speech foundation large model SparkAudio 1.0 has launched an API service on the iFlytek Open Platform, and developers can directly integrate and call it.
SparkAudio 1.0 supports input in two modalities, text and speech, and output in the text modality. It can support the implementation of scenario tasks such as speech transcription, multilingual translation, multi-dialect recognition, environmental sound understanding, speaker separation, sentiment analysis, and complex audio question answering.
According to the official introduction, in the past, audio processing mostly used a cascaded approach: first convert speech to text, then hand it to a large model for understanding. This approach has two obvious shortcomings:
- Information loss: Text can only record what was said, and loses key information inherent in speech such as tone, emotion, and background environmental sound, resulting in incomplete judgment of the scene by the model, which naturally also affects the final result.
- Fragmented pipeline: This system splits capabilities such as speech transcription, voiceprint recognition, and speech translation into independent modules run in series. There are breakpoints between modules, which not only easily accumulates errors but also causes latency and lag in the user experience.
According to the introduction, the speech foundation large model solves these problems: it no longer only does speech-to-text, but directly understands speech. The speech foundation large model understands human voice, environmental sound, music, and more in one pass, and can also understand the semantics, emotion, and scene in the sound.
Compared with the previous-generation model, the Spark speech foundation SparkAudio 1.0 has new upgrades in three major aspects:
- Hear more: supports 99 languages and 202 dialects, and can handle multilingual speech recognition and translation tasks;
- Understand more deeply: from speech recognition to multi-dimensional audio understanding, it can understand the deeper meaning of speech and supports many scenarios such as intelligent customer service, automotive cockpits, cross-language meetings, medical records, and content moderation;
- Use more broadly: trained on fully domestic computing power, providing developers and industry applications with an independently controllable speech intelligence foundation.
- Experience link: https://iflytekresearch.iflytek.com/console/spark-audio
- API address: https://www.xfyun.cn/services/Spark_Audio_1.0
- “iFlytek launches the first Spark speech foundation large model SparkAudio 1.0: switch freely between Chinese and English, real-time speaker recognition, and usable right in the input method”
Advertisement statement: External redirect links contained in this article (including but not limited to hyperlinks, QR codes, passcodes, etc.) are used to convey more information and save selection time. The results are for reference only. All articles on IT Home containing external links include this statement.
Related articles
- NASA's two probes complete special Earth gravity slingshot maneuver, officially head to Mars
- China's Long March 12 rocket successfully launches 29 groups of low-orbit satellites for satellite internet
- China successfully launches Yaogan-51 07-09 satellites, mainly used for scientific experiments, electromagnetic environment detection, and other fields
- New discovery by ESA probe: Martian volcanoes may be dormant, not geologically dead
- NASA simulated Mars mission experiment ends; four astronauts exit after 378 days of isolation
- iFlytek