MathSpeech
收藏资源简介:
MathSpeech是由首尔国立大学开发的一个用于评估自动语音识别(ASR)模型在数学语音识别能力上的基准数据集。该数据集包含1101个从YouTube上的数学讲座录音中提取的音频样本,旨在解决当前ASR模型在处理数学表达式时的性能不足问题。数据集的创建过程包括从公开的数学讲座视频中提取音频,并通过特定的处理方法生成用于训练和评估的数据。MathSpeech数据集主要应用于数学教育领域,旨在通过提高数学语音到公式转换的准确性,改善学习者的理解效果。
MathSpeech is a benchmark dataset developed by Seoul National University to evaluate the performance of automatic speech recognition (ASR) models on mathematical speech recognition tasks. This dataset comprises 1101 audio samples extracted from publicly available math lecture recordings on YouTube, and is designed to address the performance limitations of current ASR models when dealing with mathematical expressions. The creation of the MathSpeech dataset involves extracting audio from math lecture videos hosted on YouTube, and generating dedicated training and evaluation data via specific processing workflows. Primarily applied in the field of mathematics education, the MathSpeech dataset aims to enhance learners' comprehension by improving the accuracy of mathematical speech-to-formula conversion.

- 1MathSpeech: Leveraging Small LMs for Accurate Conversion in Mathematical Speech-to-Formula首尔国立大学 · 2024年



