Parcollet Titouan

SpeechLLM and Whisper

7

### Describe the bug This issue serves as a placeholder for the introduction of two recipes: full whisper pretraining and a speechLLM finetuning. People potentially interested are added. Anyone can...

SpeechBrain and ONNX

7

### Describe the bug This issue serves as a placeholder to hold the discussion and progress of tracking the compatibility of SpeechBrain with ONNX. We have many issues / discussions...

### Describe the bug This issue serves as a placeholder for the introduction of SpeechBrain pipelines. The idea is to add a repository or a directory showcasing complex pipelines (SpeechLM,...

Fix issue with weight loading for torch 2.6

5

This is fixing #2856 . This error may happen in other places, but i don't think we should fix them all as they should be considered individually to avoid introducing...

SpeechLLM (with LLaMA) and Conformer recipe for speech translation on CoVoST (Code from Samsung AI Center Cambridge)

1

This PR introduces two speech translation recipes: one training a conformer encoder decoder with MT training (ASR+ST) from scratch. The other finetuning an XLS-R into an adapter before a LLama...

25,000 of diverse English ASR data (dataset name hidden) (code from Samsung AI Center Cambridge)

5

This PR introduces the data preparation of **[insert hidden name]**(anonymity due to conference rules for now). The 6 subsets are (more details in the readme): 1. **large** contains 25,000 hours...

Why is wav2vec 2.0 masking prob set to 0.65?

1

## ❓ Questions and Help Hi there! Quick question on the masking of wav2vec 2.0. In the original paper, it is mentioned that each frame has a p=0.065 chance of...

### System Info Hello there, I'm fine-tuning a Llama 3 model from HuggingFace with PeFT and BitsAndBytes. Interestingly, when wrapping the model with DDP, the training end up taking more...