# Unormalized AMI

```python
from datasets import load_dataset

ami = load_dataset("speech-seq2seq/ami", "ihm")
```

## TODO(PVP) - explain exactly what normalization was accepted what wasn't