Open pabloapast opened 6 years ago
Hello, It depends on multiple factors such as language and the dialect (e.g. en-GB, de-CH, fr-CA) in which you are trying to recognise, the nature of speech (because a model trained using isolated words may not work well on conversational or telephone speech), the level of background noise, reverberation, sampling rate etc. Current speech recognition systems work well when the above parameters (some strictly) match between training and testing. So it's just easier to pick a dataset with similar conditions as you expect during testing and train the models. If you are interested and are familiar with ASR, I can share nnet3 DNN-HMM hybrid models trained on AMI corpus using recordings from the independent headset microphone. But if you are starting new, I suggest that you try simpler tasks such as TIDIGITS or TIMIT (I don't have models trained for them now).
Hey, do you have the pretrained DNN only model on keras for TIMIT dataset. Could you also share for DNN-HMM model ? Thanks in advance..
Hi Kumar I'm learning about speech recognition and I want to know if there is any pretrained model that I can download to test it?