Hello,
I was working on an Audio training model for keyword spotting.
Looking for a Japanese Keyword audio dataset.
Please guide me by providing reference links to the dataset.
Thank you
1 Like
Hello ShashankR,
I doubt that you will find good-quality audio training in Japanese on Hugging Face. Japanese communication giants such as Softbank, NTT, and Docomo are keeping their datasets private.
Maybe you should ask around on Japanese websites such as:
Good luck!
1 Like
Hello AkimfromParis,
Thank You for the quick reply. I will note that point.
But in Hugging Face there are Japanese audio datasets, but the datasets available are of conversions between two people or a monolog.
As per your guidance, I will check Japanese websites.
Thank you
1 Like
Another option would be to use a language model capable of generating japanese to build a synthetic dataset in text which you can then convert to audio using text to speech models(caveat they would need to produce correct linguistic audio). This would allow for a high quality dataset that is targeted for your end goal. You can add diversity of audio by using different voice generators and add noise so your model can learn to generalize instead of copy.
1 Like