KoSta Tres Amigos
I'm trying to understand according to what I see in the program's behavior.
The Lyrics are calculated separately from the Stems?
Are they calculated simultaneously?
Are they calculated immediately after the Stems?
I notice that sometimes when I load the song, I click play and until it reaches the first Lyrics they have been calculated and I see the Karaoke normally.
Other times the video freezes and only shows the small waiting circle until it calculates.
And some other times I skip directly to Lyrics and they play immediately, without even a second, which means that they are already calculated somehow. This is the first time I have loaded these songs.
I have the feeling that because they are some well-known songs, maybe they have already been calculated by other users of the program and exist in some central VirtualDJ database and it pulls them directly from the Internet.
I am confused about the process.
Stems speed I have is about 6x-7x
The Lyrics are calculated separately from the Stems?
Are they calculated simultaneously?
Are they calculated immediately after the Stems?
I notice that sometimes when I load the song, I click play and until it reaches the first Lyrics they have been calculated and I see the Karaoke normally.
Other times the video freezes and only shows the small waiting circle until it calculates.
And some other times I skip directly to Lyrics and they play immediately, without even a second, which means that they are already calculated somehow. This is the first time I have loaded these songs.
I have the feeling that because they are some well-known songs, maybe they have already been calculated by other users of the program and exist in some central VirtualDJ database and it pulls them directly from the Internet.
I am confused about the process.
Stems speed I have is about 6x-7x
Posted 6 days ago @ 4:26 pm
kradcliffe
KoSta Tres Amigos wrote :
I have the feeling that because they are some well-known songs, maybe they have already been calculated by other users of the program and exist in some central VirtualDJ database and it pulls them directly from the Internet.
That is correct.
Posted 6 days ago @ 4:56 pm
KoSta Tres Amigos
kradcliffe wrote :
That is correct.
KoSta Tres Amigos wrote :
I have the feeling that because they are some well-known songs, maybe they have already been calculated by other users of the program and exist in some central VirtualDJ database and it pulls them directly from the Internet.
That is correct.
And I'm talking for Greek songs
Posted 6 days ago @ 5:12 pm
PhantomDeejay
Doesn't matter if they are Greek or not.
If the AI model has analyzed a track for Lyrics before, then the result is cached and pulled via internet.
If however the track has not been analyzed by the AI model before, then it will get analyzed at that very moment (assuming you are a PRO user and you have stems enabled).
If the AI model has analyzed a track for Lyrics before, then the result is cached and pulled via internet.
If however the track has not been analyzed by the AI model before, then it will get analyzed at that very moment (assuming you are a PRO user and you have stems enabled).
Posted 5 days ago @ 7:59 am
KoSta Tres Amigos
PhantomDeejay wrote :
Doesn't matter if they are Greek or not.
If the AI model has analyzed a track for Lyrics before, then the result is cached and pulled via internet.
If however the track has not been analyzed by the AI model before, then it will get analyzed at that very moment (assuming you are a PRO user and you have stems enabled).
If the AI model has analyzed a track for Lyrics before, then the result is cached and pulled via internet.
If however the track has not been analyzed by the AI model before, then it will get analyzed at that very moment (assuming you are a PRO user and you have stems enabled).
yes I understand..
I said Greek to show that is's not the most famous language songs like Enlish.
But so many factors here..
A track is identified "same" even if it has a slightly different duration as a file? As long as the song is exactly the same duration?
Also, the process used by the program means that for each song, the first analysis by a user is enough? And will everyone else find it ready? Or how many times is the same song likely to be analyzed?
Also, how will any errors be corrected? If the same user re-analyzes?
If a user manually corrects an error in the lyrics, is it also updated in the company's database? And if the user's correction is wrong?
It would be interesting to read somewhere the details of how the whole system works.. If anyone knows a source, let us know
Posted 5 days ago @ 11:27 pm
PhantomDeejay
All these questions have been answered here on the forum when the feature was announced and released.
Under the hood VirtualDJ now calculates an "audio signature id" for each song.
This audio signature takes a lot of things into account in order to get calculated, and it also requires stems.
This audio signature is extremely good at being able to identify "same" songs despite duration being somehow different, or minor variations in tempo (for songs recorded by analog media), having different bitrate and/or samplerate .e.t.c.
This means that it's capable to detect when a song is the same with another with a confidence level of about 99% !!!
When you ask for an AI analysis of the lyrics, VirtualDJ passes this audio signature ID to the AI model among other information.
That's how the AI can determine if it has done an analysis or not for a given track.
After the first analysis is done for a song, a cached result gets stored.
Once it gets stored it doesn't matter how many users ask for it or if they will reanalyze or edit the result.
All the users will initially get served that first cached result, and if it's wrong they can choose to reanalyze or edit.
Edits are not reuploaded back on the AI. This is a security measure to ensure that a rogue user will not start pranking others by messing with the lyrics.
Only the VirtualDJ team has access to "change/reset" the cached result (e.g. if it sees that a significant amount of users requests a reanalyze of a particular song) and still only under certain circumstances.
Finally keep in mind that for some of the steps explained above it's not possible to provide further details as they are protected proprietary functions that Atomix would not like to disclose to their competitors.
Therefore, for instance, please don't ask for more details on how the "audio signature id" gets calculated.
You won't receive a more detailed answer.
Thank you
Under the hood VirtualDJ now calculates an "audio signature id" for each song.
This audio signature takes a lot of things into account in order to get calculated, and it also requires stems.
This audio signature is extremely good at being able to identify "same" songs despite duration being somehow different, or minor variations in tempo (for songs recorded by analog media), having different bitrate and/or samplerate .e.t.c.
This means that it's capable to detect when a song is the same with another with a confidence level of about 99% !!!
When you ask for an AI analysis of the lyrics, VirtualDJ passes this audio signature ID to the AI model among other information.
That's how the AI can determine if it has done an analysis or not for a given track.
After the first analysis is done for a song, a cached result gets stored.
Once it gets stored it doesn't matter how many users ask for it or if they will reanalyze or edit the result.
All the users will initially get served that first cached result, and if it's wrong they can choose to reanalyze or edit.
Edits are not reuploaded back on the AI. This is a security measure to ensure that a rogue user will not start pranking others by messing with the lyrics.
Only the VirtualDJ team has access to "change/reset" the cached result (e.g. if it sees that a significant amount of users requests a reanalyze of a particular song) and still only under certain circumstances.
Finally keep in mind that for some of the steps explained above it's not possible to provide further details as they are protected proprietary functions that Atomix would not like to disclose to their competitors.
Therefore, for instance, please don't ask for more details on how the "audio signature id" gets calculated.
You won't receive a more detailed answer.
Thank you
Posted 4 days ago @ 9:16 am
KoSta Tres Amigos
PhantomDeejay wrote :
All these questions have been answered here on the forum when the feature was announced and released.
Under the hood VirtualDJ now calculates an "audio signature id" for each song.
This audio signature takes a lot of things into account in order to get calculated, and it also requires stems.
This audio signature is extremely good at being able to identify "same" songs despite duration being somehow different, or minor variations in tempo (for songs recorded by analog media), having different bitrate and/or samplerate .e.t.c.
This means that it's capable to detect when a song is the same with another with a confidence level of about 99% !!!
When you ask for an AI analysis of the lyrics, VirtualDJ passes this audio signature ID to the AI model among other information.
That's how the AI can determine if it has done an analysis or not for a given track.
After the first analysis is done for a song, a cached result gets stored.
Once it gets stored it doesn't matter how many users ask for it or if they will reanalyze or edit the result.
All the users will initially get served that first cached result, and if it's wrong they can choose to reanalyze or edit.
Edits are not reuploaded back on the AI. This is a security measure to ensure that a rogue user will not start pranking others by messing with the lyrics.
Only the VirtualDJ team has access to "change/reset" the cached result (e.g. if it sees that a significant amount of users requests a reanalyze of a particular song) and still only under certain circumstances.
Finally keep in mind that for some of the steps explained above it's not possible to provide further details as they are protected proprietary functions that Atomix would not like to disclose to their competitors.
Therefore, for instance, please don't ask for more details on how the "audio signature id" gets calculated.
You won't receive a more detailed answer.
Thank you
Under the hood VirtualDJ now calculates an "audio signature id" for each song.
This audio signature takes a lot of things into account in order to get calculated, and it also requires stems.
This audio signature is extremely good at being able to identify "same" songs despite duration being somehow different, or minor variations in tempo (for songs recorded by analog media), having different bitrate and/or samplerate .e.t.c.
This means that it's capable to detect when a song is the same with another with a confidence level of about 99% !!!
When you ask for an AI analysis of the lyrics, VirtualDJ passes this audio signature ID to the AI model among other information.
That's how the AI can determine if it has done an analysis or not for a given track.
After the first analysis is done for a song, a cached result gets stored.
Once it gets stored it doesn't matter how many users ask for it or if they will reanalyze or edit the result.
All the users will initially get served that first cached result, and if it's wrong they can choose to reanalyze or edit.
Edits are not reuploaded back on the AI. This is a security measure to ensure that a rogue user will not start pranking others by messing with the lyrics.
Only the VirtualDJ team has access to "change/reset" the cached result (e.g. if it sees that a significant amount of users requests a reanalyze of a particular song) and still only under certain circumstances.
Finally keep in mind that for some of the steps explained above it's not possible to provide further details as they are protected proprietary functions that Atomix would not like to disclose to their competitors.
Therefore, for instance, please don't ask for more details on how the "audio signature id" gets calculated.
You won't receive a more detailed answer.
Thank you
OK! thanks for the explanation!
"audio signature id" is like Shazam and related programs work I think
So it is a good approach
Posted 4 days ago @ 2:51 pm