2 ms·
Donate your voice for Timers and Such
Hello HN!
I'm creating a new open-source dataset for speech research. We'd like to record people saying commands involving numbers, for things like setting timers (hence the name "Timers and Such").
Donate your voice if you have 5 minutes and want to help out this humble Ph.D. student!
https://www.timersandsuch.com
Fun fact: two participants now have independently told me that they found doing this "oddly satisfying". So if not for the sake of reproducible speech recognition research, do it for the chance for some odd satisfaction!
(If you want to learn more about the motivation for the project and why more data would help, I wrote a blog post about a preliminary version of the dataset, made using the voices of some friends and colleagues: https://lorenlugosch.github.io/posts/2020/12/slu/)
- yorwba 6y agoIs there a reason this isn't just a subset of the Mozilla Common Voice dataset, similar to their "Single Word Target Segment"? (Described as "a use case driven segment containing data to power spoken digit recognition and yes / no detection.") https://commonvoice.mozilla.org/en/datasets https://commonvoice.mozilla.org/en/datasets
- m_nemo_syne 6y agoI don't think Common Voice contains the type of commands we're recording here. Also, the numbers in these commands are like "19.56", as opposed to single digits.
- yorwba 6y agoWhat I mean is that those commands could be added to the pool that Common Voice asks users to record.
- m_nemo_syne 6y agoThat's a good idea. I can look into it.
- m_nemo_syne 6y ago(though: Common Voice is great, and you should contribute to them, too!)