Challenges

  • Click on ‘Automatic Speech Recognition’ and then on ‘1st Challenge Leadearboard’, check the ‘Closed task’ results

  • 1st of 13 academic and industry participants

  • Click on ‘Text to Speech’ and then on ‘Hindi TTS Leadership Board’

  • 1st in both naturalness and intelligibility of synthesized speech.

Multi-source sound event detection in real life conditions, where the events can occur in both isolation or overlapped. The performance of systems were evaluated based on the error rate, which is zero for an ideal system. The submitted system performed the best among 13 competitors in the challenge.

[Results] [Submitted system description] [Code]

Sharath Adavanne and Tuomas Virtanen in Detection and Classification of Acoustic Scenes and Events (DCASE 2017)

Large-scale detection of sound events using weakly labeled training data. The performance of systems were evaluated based on the error rate, which is zero for an ideal system. The submitted system fared fifth of eight competitors in the challenge. The drop in performance was because our method was performing SED at higher resolution (40ms) while the top performing methods were using low resolution (1 sec).

[Results] [Submitted system description]

Sharath Adavanne and Tuomas Virtanen in Detection and Classification of Acoustic Scenes and Events (DCASE 2017)

Detecting bird sounds in audio is an important task for automatic wildlife monitoring. In this task, given a short audio recording, a binary decision for the presence/absence of bird sound has to be made. Two systems were submitted, which fared second and fifth among 30 competitors in the challenge.

[Results] [Submitted system 1 description] [Submitted system 2 description]

Sharath Adavanne, Emre Cakir, Konstantinos Drossos, Giambattista Parascandolo, and Tuomas Virtanen

Multi-source sound event detection in real life conditions, where the events can occur in both isolation or overlapped. The performance of systems were evaluated based on the error rate, which is zero for an ideal system. The submitted system performed the best among 17 competitors in the challenge.

[Results] [Submitted system description]

Sharath Adavanne, Giambattista Parascandolo, Pasi Pertila, Toni Heittola and Tuomas Virtanen in Detection and Classification of Acoustic Scenes and Events (DCASE 2016)

Blind separation of singer's voice from pop recordings.The submitted system fared in the top systems with good signal-to-distortion ratio for both voice and residual.

[Results] [Submitted system description]

Preeti Rao, Nagesh Nayak and Sharath Adavanne in Music Information Retrieval Evaluation eXchange (MIREX 2014)