Howdy, Stranger!

It looks like you're new here. If you want to get involved, click one of these buttons!

Consistency with output

LandarLandar Member
edited September 2014 in ScoreCloud 4 / Studio
After some experiementation, I have been finding that the exact same microphone input never produces the same result twice. Sometimes the result is drastically different from the input. As of yet I can find no pattern, but I would expect that provided the exact same input that I would receive consistent and identical outputs.

In order to ensure an identical input each time, I used multiple recordings and played them back through my microphone.

I tested this with:

1) a youtube video of a solo performance of Debussy's "Syrinx" for flute

2) a playback of a snippet of my copy of Anthony Braxton's _For Alto_ (solo sax - no other instrument)

3) a recording of my canary singing solo (Found here: - I played back the last 12 seconds each time [starting at 1:26] as that section contains no sound but the canary).

In each iteration I received different results - sometimes quite drastic.

I uploaded 7 trials of the canary song to my ScoreCloud account under the collection Canackett (note sure how to share it). Those 7 iterations - named Lamia 1 - 7 - are quite different from each other.

I did not save the iterations from the first two items, but they were even more different to the point of being entirely unidentifiable from the source with just one iteration of "Syrinx" actually being an almost accurate portrayal of the source material.

The lack of consistency issues are both rhythmic and harmonic.


  • JohanRJohanR Administrator
    edited September 2014

    Thank you for reaching out!

    The difference in interpretation can have several reasons, for example the microphone position could make different overtones and noise louder, leading to the analysis weighting notes differently. And if the timings of rhythms and beat is vague (witch it is with the bird), this could be enough to "guess" the first down beat as different notes each time, leading to drastically different interpretations of the rest of the piece. The same goes for the key signature.

    We'll take a look at your collection and get back to you!

    / JohanR, ScoreCloud
    ScoreCloud Staff and Mandolinist
  • Thanks for writing back!

    I am very excited by the potential for this software - and bird song transcription (this would be so much easier than doing this by ear) is one of many applications I see for it.

    I am primarily a wind instrumentalist and in addition to my bird song projects the other primary use I hope to gain from your software is to save me enormous time to transcribe as I "doodle". It would allow me such freedom to not have to stop and scribble/click into software notes and instead go back through afterwards and explore my clips for the idea seeds - with some amount of rough scoring already accomplished! If I could get this software to work for me I cannot wait to give you my money (I would love it if you offered a lump sum "lifetime" all access local and cloud option).

    I did note when testing with the recording of "Syrinx" that if I did not put the speaker literally less than an inch from my mic that the software was barely picking up notes at all - seeming to only get the loudest notes.

    As a related aside, I experimented with different mic gain setups where in other studio applications (Reason, Tracktion, Audacity) and other voice chat applications (ventrilo, Steam voice chat, voice chat in the game Firefall) it picked up perfectly.

    I tried a number of mics as well. The mic built into my computer, a quality voice chat mic, and a C1000S. The results were about the same with the voice chat and the condensor mic leading me to think that there is perhaps some tooling that could be done in the software or my computer setup as it relates to the software (I am using Windows 8.1). The mic built into my computer seems to be not so great (it failed tests across the board), so the poor results there were expected. This is troubling to me, as I would like to be able to play whatever instrument or sing/hum or mic my bird in an unobtrusive manner or have an amp in the middle of the room or what have you without having to so closely mic the software.

    Once I had the less than an inch requirement figured out, I did rig a temporary setup to ensure that the mic and the speaker were always the exact same distance and position from each other. As such, the bird song recordings are consistent in this manner.

    You are correct that due to my playback at a very slightly different starting point each time in the case of the bird video that the key signature would be interpreted differently. However, I would expect to have the same notes and rhythms picked up each time.

    In some recordings, when the canary dipped down in registers for an extremely brief moment, the software drastically overinterpreted this as something in the very low bass register. Sometimes the software never picked it up at all. Sometimes it picked up artifacts that just were not there - it seemed that the trills and the sharp note bends confused it. This was much worse in the For Alto recordings - which troubles me as solo freeform of traditional instruments seems to be one of the things this software is designed for.

    The rhythm is very important to pick up and I am not seeing these rapid figures consistently at all - certainly never even close to true to the source when the software takes a stab at it. So often I just see the first note and then rests or it being treated as a series of whole notes. The bird is very close to something I could play on various woodwinds, and the recording of "Syrinx" is a classic standard which I thought would be a good benchmark - especially since I think I remember your demo videos specifically showing someone playing flute into the software.

    After you get a chance to look at my samples, please let me know what I can do to tweak my computer setup or my recording setup to provide some better test runs.
Sign In or Register to comment.