search

dpwe / audfprint

Landmark-based audio fingerprinting
MIT License
544 stars 122 forks source link

Low common hash value yielding high match time #62

Open StudentSA opened 5 years ago

StudentSA commented 5 years ago

Hi Dan,

Firstly, the work you have done here is exceptional, Thanks.

I have a found a case where matching a recording against multiple references generates unexpected match data. Reference audio 1 (48seconds) was fingerprinted to 6654 hashes Reference audio 2 (53seconds) was fingerprinted to 7861 hashes

The resultant match data for Reference 1 indicated 1998 common hashes with a match time 41.4 seconds which is a correct detection The match data for Reference 2 indicates a common hash of only 77 but a 42.4 second match time which is an incorrect detection

I base my match decision on match time as I believed it to be the best indication of a successful match. Would you be able to advise why such a low common hash would yield such a high match time?

Thanks.

dpwe commented 5 years ago

Match duration is just the outer limits of the earliest and latest matching and aligned landmarks. Match hash density (total matching hashes divided by match time) is the most consistent measure of quality (subject to some minimum in the denominator), but raw match count is usually ok. 77 matches is a lot for a false alarm, but it’s clearly much less than 1998.

DAn.

On Wed, May 22, 2019 at 07:23 StudentSA notifications@github.com wrote:

Hi Dan,

Firstly, the work you have done here is exceptional, Thanks.

I have a found a case where matching a recording against multiple references generates unexpected match data. Reference audio 1 (48seconds) was fingerprinted to 6654 hashes Reference audio 2 (53seconds) was fingerprinted to 7861 hashes

The resultant match data for Reference 1 indicated 1998 common hashes with a match time 41.4 seconds which is a correct detection The match data for Reference 2 indicates a common hash of only 77 but a 42.4 second match time which is an incorrect detection

I base my match decision on match time as I believed it to be the best indication of a successful match. Would you be able to advise why such a low common hash would yield such a high match time?

Thanks.

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/dpwe/audfprint/issues/62?email_source=notifications&email_token=AAEGZUN7VMW4AKBAGZRUZI3PWUULNA5CNFSM4HOTJCI2YY3PNVWWK3TUL52HS4DFUVEXG43VMWVGG33NNVSW45C7NFSM4GVF4SYQ, or mute the thread https://github.com/notifications/unsubscribe-auth/AAEGZUK57XXKQTNCPXIRGLTPWUULNANCNFSM4HOTJCIQ .