This paper reports the participation of Inria to the audiobasedbird species identication challenge of LifeCLEF 2014 campaign.Inspired by recent works on ne-grained image classication, we introducean instance-based classication scheme based on the dense indexingof MFCC features and the pruning of the non-discriminant ones. To makesuch strategy scalable to the 30M of MFCC features extracted from thetens of thousands audio recordings of the training set, we used highdimensionalhashing techniques coupled with an ecient approximatenearest neighbors search algorithm with controlled quality. Further improvementsare obtained by (i) using a sliding classier with max pooling(ii) weighting the query features according to their semantic coherence(iii) making use of the metadata to lter incoherent species. Results showthe eectiveness of the proposed technique which ranked 3rd among the10 participating groups.