VoxPopuli

Download the desired language subsets using the provided scripts.

Preprocess the raw audio using the segmentation scripts.

Train an ASR model using the transcribed data and the unlabelled data for semi-supervised learning.

Evaluate the ASR model on a held-out test set.

Download the accented English speech data subset.

Train an ASR model specifically on the accented speech data.

Fine-tune a pre-trained ASR model using the accented speech data.

Evaluate the ASR model on an accented speech test set.

Download the speech-to-speech interpretation data for the desired language pairs.

Preprocess and align the source and target audio.

Train an end-to-end speech translation model.

Evaluate the speech translation model on a held-out test set.

Train an ASR model on a high-resource language using VoxPopuli.

Fine-tune the pre-trained model on a low-resource language using limited transcribed data.

Evaluate the performance on the low-resource language.

Download speech data across multiple languages.

Extract speaker embeddings from the audio.

Analyze the distribution of speaker embeddings across different demographic groups.

Develop methods for mitigating biases in speech data.