MediaSpeech
收藏资源简介:
MediaSpeech是一个开放源代码的自动语音识别(ASR)系统评估数据集,包含西班牙语、法语、土耳其语和阿拉伯语四种语言,总计10小时的数据。该数据集从相应语言的官方YouTube频道收集,并经过人工转录,错误率低于5%。创建过程中,使用了Vosk语音识别工具包进行音频分割,并通过人工转录确保数据质量。该数据集主要用于评估和比较不同ASR系统的性能,特别是在媒体领域的应用。
MediaSpeech is an open-source evaluation dataset for automatic speech recognition (ASR) systems. It covers four languages: Spanish, French, Turkish and Arabic, with a total of 10 hours of audio data. The dataset is collected from official YouTube channels of the corresponding languages, and undergoes manual transcription with an error rate below 5%. In the creation process, the Vosk speech recognition toolkit was utilized for audio segmentation, while manual transcription was adopted to guarantee data quality. This dataset is primarily intended for evaluating and comparing the performance of various ASR systems, particularly for media-related applications.
- 1MediaSpeech: Multilanguage ASR Benchmark and DatasetNTR实验室 · 2021年



