See how Deepgram stacks up. Check out our ASR Comparison Tool. ๐ŸŽ๐ŸŠ

All Posts

HamPanda: You get Hams for Playing and Pandas for Winning

Why we made HamPanda

Yeah it’s ugly … and a lot of people ask us, “Why did you make this crazy thing?”. Or “umm, What does ๐Ÿ– have to do with ๐Ÿผ?”. DOES IT MATTER? … ok wait. I can help on the first one. Why we made HamPanda …

There’s been massive progress in images

Image analysis has gotten so good recently. A lot of that has to do with good tools (deep neural networks), but there’s a significant part that’s due to large standardized image datasets that are available to train on (check ’em out).

But progress is slow for Audio

At Deepgram, we use A.I. to analyze audio. We create our own datasets and train our own models, but we recognize that there still isn’t enough in the world. Labeling audio data isn’t as fun as looking at images and video but we’re changing that.

HamPanda Makes Audio Labeling Awesome

Audio labeling is fun now. If you don’t believe me, just scroll up and check out the sweet video. If you still aren’t convinced, just try it already.

However, audio labeling sucks if you just listen to an hour of someone speaking and try to type it out. That’s not how HamPanda works. It’s a game where audio is broken into short snippets of a few seconds each. Then you (the player!) listen to the clip and write the words down. That’s not the best part.

You get Ham when you play and Pandas when you win

It’s simple. Ham is what you get when you submit a transcription. Pandas are what you get when your transcription is GOOOOD. Level up and beat your friends. Brag about your HamPanda ultra wizard level. Win life.

Don’t sweat it if your transcription doesn’t get pandas right away, you’ll probably get points later when it gets fully verified! We use crazy algorithms to determine what a good transcription is and we learn along the way.

Deep Learning Code and Datasets

Within a week, we’ll be releasing a dataset and code to train your own speech recognition engine with. The code will be found in kur and the dataset is nearly ready to be released. The world still needs more good audio data so we’ll keep HamPanda rolling it out.

Shh, don’t tell anyone yet. But we’re putting on the world’s first amazing Deep Learning Hackathon at the end of March 2017 in San Francisco. Secret Link:

come join us if you’ve got skillzz, we’ll bring the GPUs.

Follow us to A.I. land

Twitter
SnapChat
Deepgram Newsletter



Apply Now

Receive up to $100,000 to use over 12 months.

Become a Partner

When you become a partner youโ€™re in good company.

Talk to Customer Success