Advice on getting speech data

[IPsoft AWS Account]

Hi everyone,

Someone in a different department here was interested in collecting recordings of people talking for a speech recognition system here. Can anyone give me an idea of how much requesters pay for that kind of work? Like, to record yourself saying and spelling a name, or something similarly short (a few seconds).

Also, it looks like some requesters use Vocaroo to collect recordings? Are any of you aware of this site or other ways people do this? (I'm googling too, but I thought I'd check in with you all...)

Thanks!

 

[TSolo315]

For something short (5-10 words) Google and Acme data collection will pay around 7-8 cents per recording.

Both of these have built in methods of collecting the voice samples without using a third party. You click a play button, record, and hit stop -- all within the HIT.

If using a third party like Vocaroo I would expect the pay to be a little higher for the extra hassle.

 

[IPsoft AWS Account]

Hmm, that makes sense. I don't know if we're set-up to do all that within the HIT itself.
Thank you!

 

[Cara]

i love voice hits!
just today, a voice collection hit was posted from acme. $3.25 for 50 short recordings done within the hit.
even if i had to go to another website for it, it is still worth that pay.
i think snips pays 1.80? for 20, recording on another website.

 

[IPsoft AWS Account]

Thanks, @Cara , more numbers is definitely helpful. Do you have any preference between doing like a handful of recordings per HIT versus something bigger like 20 or 50? It looks like we might have the set-up to do recording within a HIT after all. And if this ends up happening, I'll do what I can to make sure it pays well.

 

[Cara]

longer ones are generally better! sometimes the system can lag in between phrases and that is frustrating.
short, single phrase hits are probably better to start with to get a feel for how it works.

 

[A6_Foul_Out]

Thanks, @Cara , more numbers is definitely helpful. Do you have any preference between doing like a handful of recordings per HIT versus something bigger like 20 or 50? It looks like we might have the set-up to do recording within a HIT after all. And if this ends up happening, I'll do what I can to make sure it pays well.

I mean you can do it survey style with a large number of recordings or jamillo style with 1 recording per hit.

Just figure out what's easiest for you and what keeps time between recordings down as to maximize recordings per hour

 

[IPsoft AWS Account]

I mean you can do it survey style with a large number of recordings or jamillo style with 1 recording per hit.

Just figure out what's easiest for you and what keeps time between recordings down as to maximize recordings per hour

Hmm okay! If I can convince the person (from the ninja department) to do this task via MTurk, we'll probably try some different stuff out to make sure it's convenient for you guys and us.