AISHELL-WakeUp-1
1561-hour Chinese and English wake-word speech database with real-home far-field and close-talk recordings
Duration
1561.12 hours
Languages
2
Sample Rate
various
Published
2019-07
Description
1Contains 3,936,003 wake-word speech recordings totaling 1561.12 hours
2Recording languages are Chinese and English, with wake words “你好，米雅” and “hi, mia”
3254 speakers participated, and the recordings were collected in China
4Recorded in real home environments with 7 recording positions
5Uses six 16-channel circular PDM microphone arrays for far-field recording and one Hi-Fi microphone for close-talk recording
6Far-field recordings are 16 kHz, 16 bit, while close-talk recordings are 44.1 kHz, 16 bit, so the sampling rate is recorded as various
7Transcribed by professional speech annotators and quality checked with 100% word accuracy
8Can be used for speaker recognition and wake-word recognition research
Language Details
| Language | Duration |
|---|---|
| Mandarin Chinese | None |
| English | None |
Publisher
License & Commercial Use
License
Apache License v2.0; Dataset Non-Open Source / requires application
Commercial Use
Non-commercial only
Resources
Paperhttps://arxiv.org/abs/1912.01231Datasethttps://opendata.aishelltech.com/wakeup-1Official Pagehttps://www.aishelltech.com/wakeup_dataSample / Demohttps://aishell-wakeup-1.oss-cn-beijing.aliyuncs.com/AISHELL-WakeUp-1-sample.zipReadmehttps://aishell-wakeup-1.oss-cn-beijing.aliyuncs.com/AISHELL-WakeUp-1.pdf