Trelis/chorus-v1-ami-2speaker-test-preds

---
license: cc-by-4.0
tags:
- predictions
- multi-speaker
- chorus
- ami
- benchmark
source_datasets:
- Trelis/ami-2speaker-test
---

# Chorus v1 — Predictions on AMI 2-Speaker Test

**Need a voice model for your domain?** Trelis builds custom ASR, TTS, and voice agent pipelines for specialist verticals (legal, medical, finance, construction) and low-resource languages. [Enquire or book a consultation →](https://trelis.com/voice-ai-services/)

Per-row predictions from [`Trelis/Chorus-v1`](https://huggingface.co/Trelis/Chorus-v1) on [`Trelis/ami-2speaker-test`](https://huggingface.co/datasets/Trelis/ami-2speaker-test). Browse in the HF viewer to play the audio and compare ref vs pred for every clip.

## Aggregate metrics

| Metric | Speaker 1 | Speaker 2 | Mean |
|---|---|---|---|
| CER | 8.58% | 10.12% | **9.35%** |
| CMER | 8.32% | 9.69% | **9.00%** |

## Schema

Each row contains:
- `audio` — the mixed 2-speaker clip
- `speaker1_ref` / `speaker1_pred` / `speaker1_cer` / `speaker1_cmer`
- `speaker2_ref` / `speaker2_pred` / `speaker2_cer` / `speaker2_cmer`
- `duration_s`, `overlap_ratio`, `loudness_db`

Sort by `speaker2_cer` descending in the viewer to surface the hardest failures first.

## License

CC-BY 4.0, inherited from the AMI source data.