AISHELL-DMASH
20000-hour Mandarin distributed microphone-array smart-home speech database
Duration
20000 hours
Languages
1
Sample Rate
None
Published
2022-04
Description
1The full name is Distributed Microphone Arrays in Smart Home (DMASH) Dataset
2Recorded in two different rooms under real smart-home scenarios
3Contains 20000 hours of speech data
4Recording devices include one close-talk microphone and seven far-field device positions
5Each far-field position includes one iPhone, one Android phone, one iPad, one microphone, and one circular microphone array with a 5 cm radius
6Includes 511 speakers, each recorded in three visits with 7 to 15 days between adjacent visits
7All speech data was transcribed by professional speech annotators and quality checked, reaching 98% text accuracy
8Designed for research and applications including speaker recognition, speech recognition, and wake-word recognition
Language Details
| Language | Duration |
|---|---|
| Mandarin Chinese | 20000 hours |
Publisher
License & Commercial Use
License
Non-Open Source / requires application
Commercial Use
Non-commercial only