---
license: cc-by-4.0
task_categories:
  - audio-classification
  - text-to-speech
language:
  - en
  - hi
  - te
  - ta
size_categories:
  - 1K<n<10K
tags:
  - speaker-encoding
  - cross-lingual
  - code-switching
  - indic
  - synthetic
---

# Codeswitch Pairs LASE — Western held-out corpus

1043 held-out cross-script utterance pairs from 8 ElevenLabs Western Multilingual voices. Used to evaluate generalisation of speaker encoders trained on `Praxel/codeswitch-pairs-lase`.

Each row is one synthesized utterance with metadata; pairs are reconstructed at evaluation time by joining on `voice_id` (same voice, different script = cross-script pair).

## Schema (manifest.jsonl)

```json
{
  "voice_id": "21m00Tcm4TlvDq8ikWAM",
  "lang": "en | hi | te | ta",
  "text": "the prompt text",
  "wav_path": "<voice_name>/<lang>_NNN.wav",
  "quality": {"cosine": 0.97, "utmos": 4.6}
}
```

Audio is 16 kHz mono WAV, ~2 s/utterance.

## Stats

- **Pairs**: 1043
- **Voices**: 8
- **Languages**: en, hi, te, ta
- **Quality gate**: WavLM-cosine ≥ 0.90 vs the voice's reference clip

## Provenance

Synthesized using the [ElevenLabs Multilingual v2](https://elevenlabs.io) API. Voice IDs are public ElevenLabs catalog voices used under their TOS for research/evaluation purposes. Text prompts are short generic English phrases translated/transliterated into each target script.

## Companion paper

LASE: Language-Adversarial Speaker Encoding for Indic Cross-Script Identity Preservation (arXiv:TBD)

## License

CC-BY-4.0.