THCHS-30
A compact collection of Mandarin read speech with aligned transcripts and companion resources for noise testing.
Visit official source (opens in a new tab)Research use
Start with the standard splits to establish a reproducible baseline, then compare recognition under the supplied noise conditions.
Limitations
The scripted recordings and noise tests cover a narrower range than spontaneous dialogue.
License and access
- Dataset license
- Apache-2.0 (opens in a new tab)
- Access status
- Direct access
Use the publisher’s resource page and check the terms for the files and version you choose.