phoenix-14T

RWTH-PHOENIX-Weather 2014T是第一个公开可用的连续手语翻译（SLT）数据集，以下是关于它的详细介绍：

### 数据来源与内容
- **来源**：该数据集是从天气预报的手语解释中获取的。
- **内容**：包含了德国手语视频、对应的口语翻译以及词汇级别的注释。视频内容主要是关于天气广播的手语表达。

### 数据规模
- **帧数**：超过95万帧。
- **手势数量**：涵盖了超过67,000个手势。
- **词汇量**：手语词汇量超过1,000个，德语词汇量超过2,800个，总计超过99,000个词汇。

### 特点与意义
- **特点**：提供了丰富的连续手语数据，且具有较为准确的口语翻译和词汇注释，为手语研究提供了高质量的语料。
- **意义**：推动了手语识别和翻译领域的研究发展，为研究人员提供了一个标准的数据集来评估和比较不同的手语识别、翻译算法和模型的性能，有助于提高手语翻译技术的准确性和实用性，促进聋人与健听人之间的交流和融合。

RWTH-PHOENIX-Weather 2014T is the first publicly available continuous sign language translation (SLT) dataset. Here is a detailed introduction to it:

### Data Source and Content
- **Source**: This dataset is collected from sign language interpretations of weather forecasts.
- **Content**: It contains German Sign Language (DGS) videos, corresponding spoken language translations, and word-level annotations. The video content mainly consists of sign language expressions for weather broadcasts.

### Dataset Scale
- **Number of frames**: Over 950,000 frames.
- **Number of sign gestures**: Covers more than 67,000 sign gestures.
- **Vocabulary size**: The sign language vocabulary exceeds 1,000 entries, the German vocabulary exceeds 2,800 entries, with a total of over 99,000 vocabulary items.

### Characteristics and Significance
- **Characteristics**: It provides rich continuous sign language data, with relatively accurate spoken language translations and word-level annotations, serving as high-quality corpus for sign language research.
- **Significance**: It has promoted the research and development in the fields of sign language recognition and translation, provides researchers with a standard dataset to evaluate and compare the performance of different sign language recognition, translation algorithms and models, helps improve the accuracy and practicality of sign language translation technologies, and facilitates communication and integration between deaf and hearing people.

搜集汇总

背景与挑战

背景概述

phoenix-14T数据集是首个公开可用的连续手语翻译（SLT）数据集，源自天气预报的手语解释，包含德国手语视频、口语翻译和词汇注释。其规模庞大，涵盖超过95万帧、67,000个手势和总计99,000个词汇，为手语识别和翻译研究提供了高质量的标准语料，推动该领域的技术发展和实际应用。