README.md

---
license: cc-by-4.0
language:
  - en
pretty_name: cond-ID audio samples
size_categories:
  - 1K<n<10K
task_categories:
  - text-to-speech
tags:
  - speech
  - text-to-speech
  - zero-shot-tts
  - voice-cloning
  - machine-unlearning
  - speaker-unlearning
  - cond-id
  - evaluation
---

# cond-ID — audio samples

Synthesized audio backing the paper **"cond-ID: Conditioning-Space Identity Redirection for Speaker Unlearning in Zero-Shot TTS."** ~2,450 clips: every backbone, every baseline, and the relearn stress test.

🔇 **Prefer listening in the browser?** → [Live demo Space](https://huggingface.co/spaces/RootAccess4Life/cond-id-demo)

## The one thing to listen for

Each speaker appears as a **pair**:

| clip | meaning |
|---|---|
| `cB_…` | **baseline** clone — the un-edited model cloning that speaker. This is the voice being copied. |
| `cE_…` | **edited** clone — the same model, same reference, *after* cond-ID unlearning. The identity should be gone. |
| `cRL_…` | **relearn** clone — after a white-box attacker re-finetunes to recover the erased speaker. |

Play `cB` then `cE` for the same speaker: intelligible speech in both, but the voice
identity should no longer match. `cRL` asks whether an adversary can undo that.

## Filename convention

```
{cB|cE|cRL}_{speaker}_{sentence}.wav
```

Speaker ids are VCTK (`p225`, `p272`, …) or LibriTTS (`103`, `118`, …) depending on the
split; `{sentence}` indexes a fixed sentence bank, so the same index is the same text
across every directory. TruS-F5 clips carry an extra seed field
(`cB_s0_p225_0.wav`).

## Directory map

**Main result — cond-ID on three backbones**

| directory | backbone |
|---|---|
| `xtts_clones/` | XTTS-v2 |
| `tortoise_clones/` | Tortoise-TTS |
| `indextts_clones/` | IndexTTS-1.5 |

**Stress tests**

| directory | what it shows |
|---|---|
| `xtts_relearn_clones/` | white-box relearn attack on XTTS-v2 (`cRL_`) |
| `sequential/{xtts,tortoise,indextts}_clones/` | sequential opt-out — speakers erased one after another |

**Baselines** (each in `{method}_{backbone}_clones/`)

`additive_gaussian` · `gradient_ascent_to_forget` · `random_reference_substitution` ·
`task_arithmetic` · `zero_pooled_identity`

**TruS on F5-TTS**

| directory | contents |
|---|---|
| `trus_f5_clones/` | TruS-F5 clones (VCTK) |
| `trus_f5_libritts_clones/` | TruS-F5 clones (LibriTTS) |
| `trus_f5_frontier/` | α/threshold frontier sweeps (JSON metrics, not audio) |

## Loading

Stream a single clip without cloning the whole set:

```python
from huggingface_hub import hf_hub_download

p = hf_hub_download(
    repo_id="RootAccess4Life/cond-id-samples",
    filename="xtts_clones/cE_103_0.wav",
    repo_type="dataset",
)
```

Or grab one backbone:

```python
from huggingface_hub import snapshot_download

snapshot_download(
    repo_id="RootAccess4Life/cond-id-samples",
    repo_type="dataset",
    allow_patterns=["xtts_clones/*"],
)
```

## Provenance and licensing

Clips are synthesized by XTTS-v2, Tortoise-TTS, IndexTTS-1.5 and F5-TTS from reference
audio in **CSTR VCTK-Corpus-0.92** (CC-BY-4.0, [DOI 10.7488/ds/2645](https://doi.org/10.7488/ds/2645))
and **LibriTTS** (CC-BY-4.0, OpenSLR SLR60). This dataset is released **CC-BY-4.0** to
match those sources.

The generating models keep their own licenses — notably **XTTS-v2 is non-commercial
(CPML)**. Those terms constrain the models, not these audio files, but review them before
building on the pipeline.

## Intended use and limits

Released for **research and demonstration**: verifying the paper's claims by ear, and
supporting work on speaker unlearning, privacy, and the right to be forgotten in speech
synthesis.

These are synthetic clones of real speakers, produced to demonstrate *removing* voice
identity. Do not use them to impersonate anyone, to train or evaluate systems that
attribute speech to the original speakers, or as evidence any of these people said these
words — **they did not**. The `cB_` clips in particular are deliberate voice clones and
should be handled accordingly.

## Related

- 💻 Code: https://github.com/pujariaditya/cond-id-tts-unlearning
- 🔇 Live demo: https://huggingface.co/spaces/RootAccess4Life/cond-id-demo
- 📦 Mirrors: [XTTS-v2](https://huggingface.co/RootAccess4Life/cond-id-xtts-v2) · [IndexTTS-1.5](https://huggingface.co/RootAccess4Life/cond-id-indextts-1.5)

## Citation

```bibtex
@misc{pujari_condid,
  title  = {cond-ID: Conditioning-Space Identity Redirection for Speaker Unlearning in Zero-Shot TTS},
  author = {Pujari, Aditya and Rattani, Ajita},
  note   = {Preprint},
}
```