---
license: cc0-1.0
language:
- en
task_categories:
- automatic-speech-recognition
pretty_name: Common Voice English Scripted Test Set 500 (Trelis)
---

# Common Voice English Scripted Test Set — 500 clips

**n = 500** utterances · private eval set for ASR benchmarking

## Source

Derived from **Mozilla Common Voice Scripted Speech 25.0 — English** (`test` split), downloaded via the [Mozilla Data Collective](https://mozilladatacollective.com/) API (dataset ID `cmndapwry02jnmh07dyo46mot`, 94 GB tarball).

## Construction

Starting from the full CV 25.0 English `test` split (16,398 rows), a stratified 500-clip subset was produced using the same recipe as `Trelis/cv-hi-test-500`:

| Step | Rule |
|---|---|
| Quality filter | `up_votes ≥ 2`, `down_votes = 0` |
| Duration filter | `0.5 s ≤ duration ≤ 25 s` |
| Speaker cap | max 4 clips per `client_id` |
| Stratification | Proportional allocation across duration buckets (0–3 s, 3–6 s, 6–10 s, 10–25 s) |
| Seed | 42 |

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

## Columns

| Column | Type | Description |
|---|---|---|
| `audio` | Audio bytes (MP3) | Speech recording |
| `transcription` | string | Reference text |
| `id` | string | CV clip path stem |
| `client_id` | string | Anonymised speaker hash |
| `duration` | float | Duration in seconds |
| `age` / `gender` / `accents` | string | Speaker metadata (may be empty) |
| `up_votes` / `down_votes` | int | Community quality votes |

## License

[CC0 1.0](https://creativecommons.org/publicdomain/zero/1.0/) — Mozilla Common Voice audio and text are released into the public domain by contributors.