AI4D Baamtu Datamation - Automatic Speech Recognition in WOLOF
$2,000 USD
Can you create an automatic speech recognition model for Wolof for use in public transport?
289 data scientists enrolled, 47 on the leaderboard
TransportationAutomatic Speech RecognitionNLPAudio
12 February—23 May
101 days
12th place solution
published 27 May 2021, 17:37

Hello everyone, this is the summary of the 12th solution:

Fine-tune facebook wav2vec2 without any data augmentation or parameter tuning, got 12% WER.

Using two effects (reduce speed + reverberation) to transform data with p=0.5, reduce both the attention_dropout and the hidden_dropout to 0.05, got 0.099 WER.

Other effects such as adding noise, gain, and pitch shift did not improve the results.