SPGISpeech

SPGISpeech 由 5，000 小时的公司收益电话会议录音及其各自的转录组成。最初的呼叫被分成长度从5到15秒不等的切片，以便于训练语音识别系统。电话代表了国际商务英语的广泛横截面;SPGISpeech包含大约50，000名发言者，是所有语音语料库中人数最多的语言之一，并提供各种L1和L2英语口音。每个 WAV 文件的格式为单通道、16kHz、16 位音频SPGISpeech consists of 5,000 hours of corporate earnings conference call recordings and their corresponding transcriptions. The original calls were segmented into clips ranging from 5 to 15 seconds in length to facilitate the training of speech recognition systems. The calls represent a broad cross-section of international business English; SPGISpeech includes approximately 50,000 speakers, making it one of the largest speech corpora in terms of speaker count, and covers a variety of L1 and L2 English accents. Each WAV file is formatted as single-channel, 16kHz, 16-bit audio.

搜集汇总

背景与挑战

背景概述

SPGISpeech是一个包含5,000小时公司收益电话会议录音及转录的数据集，音频格式为单通道、16kHz、16位，涵盖约50,000名发言者的国际商务英语及多种口音，由英伟达·Kensho Technologies于2021年发布。