# How does Sound Hound work?

**URL:** https://boards.straightdope.com/t/how-does-sound-hound-work/672489
**Category:** Factual Questions
**Created:** [October 28, 2013, 9:47pm UTC](https://boards.straightdope.com/t/how-does-sound-hound-work/672489 "2013-10-28T21:47:20Z")
**Posts on this page:** 5
**Page:** 1

<div class="post-metadata">

### Author: ![Bijou\_Drains](https://sea3.discourse-cdn.com/straightdope/user_avatar/boards.straightdope.com/bijou_drains/32/2814_2.png) [@Bijou\_Drains](https://boards.straightdope.com/u/Bijou_Drains)
#### Post date: [October 28, 2013, 9:47pm UTC](https://boards.straightdope.com/t/how-does-sound-hound-work/672489/1 "2013-10-28T21:47:20Z")

</div>

It’s pretty amazing how fast it can ID a song. I assume the database it checks is not on my phone. It makes a few errors. I guess it’s looking at a digital signature of the song?

FYI , it’s an Android app that can listen to a song and find the title and artist. It works on obscure songs too, not just hits. And it takes about 5 seconds to find the song.

---

<div class="post-metadata">

### Author: ![friedo](https://avatars.discourse-cdn.com/v4/letter/f/8edcca/32.png) [@friedo](https://boards.straightdope.com/u/friedo)
#### Post date: [October 28, 2013, 9:54pm UTC](https://boards.straightdope.com/t/how-does-sound-hound-work/672489/2 "2013-10-28T21:54:48Z")

</div>

Assuming it works similarly to Shazam, the basic idea is to take a [FFT](http://en.wikipedia.org/wiki/FFT) of the audio, upload that to a server somewhere, have the server use various statistical methods to compare the FFT to known FFTs of zillions of songs, mix in a little secret sauce, and return the result to your phone.

---

<div class="post-metadata">

### Author: ![ekedolphin](https://sea3.discourse-cdn.com/straightdope/user_avatar/boards.straightdope.com/ekedolphin/32/10003_2.png) [@ekedolphin](https://boards.straightdope.com/u/ekedolphin)
#### Post date: [October 29, 2013, 12:01pm UTC](https://boards.straightdope.com/t/how-does-sound-hound-work/672489/3 "2013-10-29T12:01:33Z")

</div>

FYI, in my experience with the app, it doesn’t work quite as well with classical songs (of which there might be hundreds of different recordings, with different instruments and in different tempos) as it does with non-classical. I say not “quite” as well because it _did_ manage to identify “Baby Elephant Walk” by Henry Mancini.

I heard it every night on NPR, didn’t recognize it and it was driving me crazy. (That’s why I got SoundHound in the first place).

---

<div class="post-metadata">

### Author: ![Lips\_Obsession](https://avatars.discourse-cdn.com/v4/letter/l/bbce88/32.png) [@Lips\_Obsession](https://boards.straightdope.com/u/Lips_Obsession)
#### Post date: [October 30, 2013, 4:29pm UTC](https://boards.straightdope.com/t/how-does-sound-hound-work/672489/4 "2013-10-30T16:29:11Z")

</div>

I wish these kind of apps would be able to identify “variants” of songs. For example, we have a local radio station that regularly plays remixes of the “standard” radio play versions. I know these are available, but I usually can’t identify them. The app I use just identifies it as the standard version.

---

<div class="post-metadata">

### Author: ![CurtC](https://avatars.discourse-cdn.com/v4/letter/c/ce73a5/32.png) [@CurtC](https://boards.straightdope.com/u/CurtC)
#### Post date: [October 30, 2013, 7:32pm UTC](https://boards.straightdope.com/t/how-does-sound-hound-work/672489/5 "2013-10-30T19:32:01Z")

</div>

> [@friedo](#):
>
> Assuming it works similarly to Shazam, the basic idea is to take a [FFT](http://en.wikipedia.org/wiki/FFT) of the audio, upload that to a server somewhere, have the server use various statistical methods to compare the FFT to known FFTs of zillions of songs, mix in a little secret sauce, and return the result to your phone.

SoundHound is a little different. My friend told me that it identified a song for him just by his humming the melody for it. So I got the app and did a fairly crappy humming of the guitar lick from “Sunshine of Your Love” by Cream, and the correct title was the very first one it suggested as a match.

I really doubt that the FFT from my crappy humming would in any way match the actual recording.
