# AMI Meeting Corpus (Kaldi-segmented ASR release)

## Facts

- Provider: AMI Consortium; University of Edinburgh CSTR
- License: CC-BY-4.0 (https://groups.inf.ed.ac.uk/ami/corpus/license.shtml)
- Size: 89 files
- Format: Parquet (in sampled files)
- Stored at: Hugging Face, https://huggingface.co/datasets/edinburghcstr/ami
- Access: public, readable without an account
- Category: Speech & Audio
- Source: https://groups.inf.ed.ac.uk/ami/corpus/
- Page: https://scigantic.com/archive/4c6a5a6d-14bb-5cd8-a90e-09a92f6044f5
- Keywords: Speech & Audio, speech recognition, benchmark, Parquet, Hugging Face

## About

The AMI Meeting Corpus consists of 100 hours of English meeting recordings made in three rooms with different acoustics, mostly with non-native speakers. The original recordings use a range of signals synchronised to a common timeline: close-talking and far-field microphones, individual and room-view video, slide projector output and an electronic whiteboard. It is a standard benchmark for meeting transcription, far-field and distant speech recognition, speaker diarization and meeting summarization research. This Hugging Face release from the University of Edinburgh's Centre for Speech Technology Research follows Kaldi's AMI s5 recipe: text is normalised and audio is chunked into utterances. It holds two configurations, ihm (individual headset microphones, 51 Parquet shards) and sdm (single distant microphone, 36 shards), each with train, validation and test splits; rows carry meeting id, audio id, text, audio, begin and end time, microphone id and speaker id (measured 2026-11-14; the Hub reports about 48 GB). Load with datasets.load_dataset('edinburghcstr/ami', 'ihm'). The AMI corpus and its annotations are released under CC BY 4.0; credit the AMI Consortium.

## Load the data

```python
from datasets import load_dataset
ds = load_dataset("edinburghcstr/ami")
print(ds)
```

## Data structure

File types in sampled files: Parquet

Example file: `ihm/test-00000-of-00004.parquet`

## Open it

- In a hosted notebook with the data mounted: https://scigantic.com/archive/4c6a5a6d-14bb-5cd8-a90e-09a92f6044f5
- Full record as JSON (schema card included): https://api.scigantic.com/api/archives/4c6a5a6d-14bb-5cd8-a90e-09a92f6044f5
- Over MCP: `get_schema_card` with id `4c6a5a6d-14bb-5cd8-a90e-09a92f6044f5` at https://api.scigantic.com/mcp

## Related topics

- [Parquet datasets](https://scigantic.com/topics/parquet)
- [Benchmark datasets](https://scigantic.com/topics/benchmark)
- [Speech & Audio datasets](https://scigantic.com/topics/speech-and-audio)
- [Speech recognition datasets](https://scigantic.com/topics/speech-recognition)
