issai/Uzbek_Speech_Corpus

乌兹别克语音语料库（USC）是一个包含958位说话者、105小时转录音频记录的开放源代码语音数据集。该数据集用于自动语音识别任务，包含DNN-HMM和端到端(E2E)两种架构的识别结果，验证集和测试集上的单词错误率分别为18.1%和17.4%。语料经过母语人士手动检查，并遵循MIT许可证发布。

The Uzbek Speech Corpus (USC) is an open-source speech dataset containing 958 speakers and 105 hours of transcribed audio recordings. It is used for automatic speech recognition tasks with recognition results for both DNN-HMM and End-to-End (E2E) architectures, showing a Word Error Rate (WER) of 18.1% on the validation set and 17.4% on the test set. The corpus has been manually checked by native speakers and is released under the MIT license.