CNN for Speaker Classification with TensorFlow and LibriSpeech
Budget: $30 – $250 USD
I need assistance with an assignment involving the design and training of a Convolutional Neural Network (CNN) for speaker classification. The model will take Mel Spectrograms of speech audio from the LibriSpeech dataset as input.
The solution should provide code for the following components:
A. Code for converting raw audio files into Mel spectrograms (I think Librosa is good for this).
B. Normalizing and resizing spectrograms to fit the CNN input.
C. Splitting the dataset into train, validation, and test sets.
D. Designing a CNN model.
E. Creating a training pipeline with an appropriate loss function (maybe CTC loss) and optimizer.
F. Writing an evaluation measure suitable for the task.
The project should utilize the Librispeech-clean-100 dataset. The solution should include suggestions on how to modify the code for raw speech input instead of spectrograms.
Please ensure the solution is explainable and resources used are mentioned in the notebook. A Jupyter notebook format would be preferred.
The solution should provide code for the following components:
A. Code for converting raw audio files into Mel spectrograms (I think Librosa is good for this).
B. Normalizing and resizing spectrograms to fit the CNN input.
C. Splitting the dataset into train, validation, and test sets.
D. Designing a CNN model.
E. Creating a training pipeline with an appropriate loss function (maybe CTC loss) and optimizer.
F. Writing an evaluation measure suitable for the task.
The project should utilize the Librispeech-clean-100 dataset. The solution should include suggestions on how to modify the code for raw speech input instead of spectrograms.
Please ensure the solution is explainable and resources used are mentioned in the notebook. A Jupyter notebook format would be preferred.