正文 · AI 翻译
Thanks to IT Home reader Domado for the tip!
IT Home reported on November 12 that iFlytek announced today that SparkAudio 1.0, its Spark foundational speech model, is now available as an API service on the iFlytek Open Platform, allowing developers to integrate and call it directly.
SparkAudio 1.0 accepts text and audio input and produces text output. It supports tasks including speech transcription, multilingual translation, recognition of multiple dialects, understanding of environmental sounds, speaker diarization, sentiment analysis, and complex audio question answering.
According to the company, audio was previously processed mostly through a cascaded approach: speech was first converted to text, which was then passed to a large model for understanding. This approach has 2 clear shortcomings:
Information loss: Text can record only what was said, losing key information carried by the audio itself, such as tone, emotion, and background sounds. This leaves the model with an incomplete understanding of the scene and naturally affects the final result.
A fragmented pipeline: This system splits capabilities such as speech transcription, voiceprint recognition, and speech translation into separate modules that run in sequence. Gaps between the modules can cause errors to accumulate, as well as delays and stuttering in the user experience.
According to the introduction, the foundational speech model addresses these problems: instead of merely converting speech to text, it understands audio directly. It can understand human voices, environmental sounds, music, and more in a single pass, while also grasping the meaning, emotion, and context within the sounds.
Compared with the previous generation, the SparkAudio 1.0 foundational speech model has been upgraded in 3 major areas:
Hears more: It supports 99 languages and 202 dialects and can handle multilingual speech recognition and translation tasks;
Understands more deeply: Moving from speech recognition to multidimensional audio understanding, it can grasp the deeper meaning of speech and supports scenarios including intelligent customer service, car cabins, cross-language meetings, medical records, and content moderation;
Wider use: Trained entirely on domestically produced computing resources, it provides developers and industry applications with an independently controllable foundation for speech intelligence.
IT Home provides the relevant links:
Try it: https://iflytekresearch.iflytek.com/console/spark-audio
API address: https://www.xfyun.cn/services/Spark_Audio_1.0
Related reading:
“iFlytek debuts SparkAudio 1.0 foundational speech model: Switch freely between Chinese and English, identify speakers in real time, and use it in the input method”
Advertising disclosure: External links in this article (including but not limited to hyperlinks, QR codes, and passcodes) are provided to convey more information and save time in evaluating options. The results are for reference only. All IT Home articles containing external links include this disclosure.