TwinkStart
commited on
Upload README.md with huggingface_hub
Browse files
README.md
CHANGED
@@ -1,29 +1,50 @@
|
|
1 |
-
|
2 |
-
|
3 |
-
|
4 |
-
|
5 |
-
|
6 |
-
|
7 |
-
|
8 |
-
|
9 |
-
|
10 |
-
|
11 |
-
|
12 |
-
|
13 |
-
|
14 |
-
|
15 |
-
|
16 |
-
|
17 |
-
|
18 |
-
|
19 |
-
|
20 |
-
|
21 |
-
|
22 |
-
|
23 |
-
|
24 |
-
|
25 |
-
|
26 |
-
|
27 |
-
|
28 |
-
|
29 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
|
2 |
+
![logo.png](assets/logo.png)
|
3 |
+
|
4 |
+
|
5 |
+
> This dataset only contains test data, which is integrated into UltraEval-Audio(https://github.com/UltraEval/UltraEval-Audio) framework.
|
6 |
+
|
7 |
+
# UltraEval-Audio: 🎙️ Open-Source Speech Model Evaluation Framework, Empowering Your AI Voice Research!
|
8 |
+
|
9 |
+
## One-Stop Evaluation, Time-Saving and Effortless! 🚀
|
10 |
+
|
11 |
+
UltraEval-Audio integrates **30+ benchmarks**, covering *speech*, *sound*, *medicine*, *music* four major domains, supporting **10 languages**, and encompassing **12 types of tasks**, helping you easily master speech evaluation!
|
12 |
+
|
13 |
+
## No Tedious Operations, Ready to Use Out of the Box! 🎁
|
14 |
+
|
15 |
+
- Automatic Download and Management of Benchmarks, Say Goodbye to Manual Download and Processing! 📥
|
16 |
+
- Built-in 8 Common Evaluation Methods, Including **wer, wer-zh, G-Eval**, Meeting Your Diverse Evaluation Needs! 🛠️
|
17 |
+
|
18 |
+
## Flexible Expansion, Seamless Integration! 🔗
|
19 |
+
|
20 |
+
- Quick Integration of Internal Datasets, Custom Dataset Functionality, Making Your Evaluation More Targeted! 📊
|
21 |
+
- Easy Access to Existing Evaluation Systems, Excellent Scalability and Standardization, Allowing UltraEval-Audio to Easily Integrate into Your Evaluation Ecosystem! 🧩
|
22 |
+
|
23 |
+
|
24 |
+
Experience Now: https://github.com/UltraEval/UltraEval-Audio 🌐
|
25 |
+
|
26 |
+
|
27 |
+
UltraEval-Audio: 🎙️ Your Best Partner in AI Voice Research, Accelerating Your Breakthroughs in Speech Technology! 🚀
|
28 |
+
|
29 |
+
|
30 |
+
# UltraEval-Audio: 🎙️ 开源语音大模型评测框架,赋能您的AI语音研究!
|
31 |
+
|
32 |
+
## 一站式评测,省时省力! 🚀
|
33 |
+
|
34 |
+
UltraEval-Audio 为您集成 **30+ benchmark**,涵盖 speech, sound, medicine, music 四大领域,支持 **10种语言**,覆盖 **12类任务**,助您轻松驾驭语音评测!
|
35 |
+
|
36 |
+
## 无需繁琐操作,开箱即用! 🎁
|
37 |
+
|
38 |
+
- 自动下载和管理 benchmark,告别手动下载和处理的烦恼! 📥
|
39 |
+
- 内置 8 种常用评测方法,包括 wer, wer-zh, G-Eval 等,满足您的多样化评测需求! 🛠️
|
40 |
+
|
41 |
+
## 灵活扩展,无缝衔接! 🔗
|
42 |
+
|
43 |
+
- 快速集成内部数据集,自定义数据集功能,让您的评测更具针对性! 📊
|
44 |
+
- 方便接入现有评测系统,良好的扩展性和规范性,让 UltraEval-Audio 轻松融入您的评测体系! 🧩
|
45 |
+
|
46 |
+
|
47 |
+
立即体验: https://github.com/UltraEval/AudioEvals 🌐
|
48 |
+
|
49 |
+
|
50 |
+
UltraEval-Audio: 🎙️ 您的 AI 语音研究最佳伙伴,助您加速语音技术突破! 🚀
|