I have been working on a machine learning project for speaker recognition. I need more audio files from different speakers to improve the accuracy of my algorithm*.

I'm using the files to recognize who is speaking. It's not a speech to text algorithm.

I'm using just digits (initially only zero) to simplify my task, because it's still a proof of concept.

*I tested my machine learning with just 10 speakers and it gave me 55% accuracy. I want to add more samples to get a higher percentage.


1 Answer 1


There are multiple datasets of spoken digits.


30000 audio samples of spoken digits (0-9) of 60 different speakers. Download. Paper

Free Spoken Digit Dataset (FSDD)

5 speakers, 2,500 recordings (50 of each digit per speaker). Download

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.