---
license: cc-by-nc-4.0
language:
- hi
- en
task_categories:
- automatic-speech-recognition
pretty_name: CS-FLEURS Hindi-English Read Test (Trelis)
---

# CS-FLEURS Hindi-English Read Test

**n = 233** utterances · private eval set for code-switched ASR benchmarking

## Source

Derived from [byan/cs-fleurs](https://huggingface.co/datasets/byan/cs-fleurs) (`read/test` split, `language = hin-eng`).

CS-FLEURS is a code-switching extension of FLEURS covering Hindi-English mixed speech.

## Construction

Filtered to `language == "hin-eng"` rows from the `read/test` split:

- Audio downloaded directly from the `byan/cs-fleurs` HF repository
- `transcription` = reference text (mixed Devanagari + Latin)
- Metadata sourced from `read/test/metadata.jsonl`

**Build script:** `merge-bench-baselines/scripts/build_cs_fleurs_hineng.py`

## Columns

| Column | Type | Description |
|---|---|---|
| `audio` | Audio bytes | Speech recording |
| `transcription` | string | Reference text (mixed script) |
| `id` | string | Original CS-FLEURS file ID |
| `language` | string | Language tag (`hin-eng`) |
| `duration` | float | Duration in seconds |

## License

[CC-BY-NC-4.0](https://creativecommons.org/licenses/by-nc/4.0/) — inherited from upstream CS-FLEURS. For internal benchmarking only; not for commercial use.