Cyleux/oc_voice_2026_word_timings

oc_voice_2026_word_timings是一个针对目标说话者电话风格话语的音频/文本数据集，包含ElevenLabs Scribe V2单词级时间戳。每行数据代表一个目标说话者/代理的对话轮次，包含对话ID、轮次索引、目标说话者标识、训练文本、ASR转录文本、文本来源、单词时间戳列表、音频数据、持续时间等信息。数据集包含10,498个对话，41,992行数据，626,600个带时间戳的单词，其中39,788行用于训练，2,204行用于验证。所有对话轮次的时长在0.5到20.0秒之间，且在导出前已排除缺失ASR、无语音、空转录或过短轮次的对话。

oc_voice_2026_word_timings is a turn-level audio/text dataset for target-speaker phone-call style utterances, with ElevenLabs Scribe V2 word-level timestamps. Each row represents one target-speaker/agent turn, containing conversation ID, turn index, target speaker identifier, training text, ASR transcript, text source, list of word timings, audio data, duration, etc. The dataset contains 10,498 conversations, 41,992 rows, 626,600 timed words, with 39,788 rows for training and 2,204 for validation. All turns are between 0.5 and 20.0 seconds, and conversations with missing ASR, [no speech], empty transcripts, or too-short turns were excluded before export.