Audio Recorder for Source Separation

We are trying to build a system which would isolate each speaker from recordings with overlapping speech. You will be asked to read the shown sentences with the help of an interface. Do NOT say arbitrary sentences. Read the shown sentences together with someone-else, while you are in the same room.

Person 1 needs to read the sentence 1, and person 2 needs to read the sentence 2. Both people should read at the same time so that we mix the two speakers. Here is an example recording which sounds like what you should record.

Sentence 1:

It is one of the earliest cultural villages yet discovered in southwest asia

Sentence 2:

Mister Icon and the investor group he had hold seventy two point seven

Each person should stay approximately the same distance from the microphone. On a given mixture, each sentence needs to be uttered by a different person. You cannot superpose your own voice, if you do that we cannot use your submission.

Read the sentences while you are in the same room, approximately same distance (not too close, not too far) from the microphone. Make sure it's possible to understand the sentences from your recording.

You can record with your laptop or telephone. No need for high quality microphones.

Before starting recording, please fill-out the form below. First choose the genders of the speakers. Then choose whether the speakers are native English speakers or not. If there are non-native speakers, please specify the mother tongue in the text box. Finally press the submit button to start your recordings.

Please try to complete your work in one session as we reset the worker ID in each session. Also, please DO NOT click the 'go back' icon of your browser during your session.

If you are working through mechanical turk, for each recording you upload do not forget to submit your work stamp on mechanical turk, in order to get paid!

Choose the genders of the speakers





Choose whether the speakers are native English speakers