# Towards Zero-Shot Attribution of Synthetic Speech via Audio-Text Contrastive Retrieval

arXiv: https://arxiv.org/abs/2609.21581 · HTML: https://mlsanity.com/paper/2609.21581/

## Key facts

- No citing papers in the map yet (data through 20 Oct 2026).
- Placed in Anti-spoofing and adversarial audio attacks, one of 11 directions in Speech and audio language models.
- Builds on 22 of its 25 references that are in the map.
- First posted on arXiv 18 Sep 2026.

## Placed in

- Speech and audio language models / [Anti-spoofing and adversarial audio attacks](https://mlsanity.com/speech/anti-spoofing-and-adversarial-audio-attacks/)

## Builds on (most cited first, 22 of 22)

- [Learning Transferable Visual Models From Natural Language Supervision](https://mlsanity.com/paper/2103.00020/) (2021)
- [Representation Learning with Contrastive Predictive Coding](https://mlsanity.com/paper/1807.03748/) (2018)
- [Scaling Up Visual and Vision-Language Representation Learning With Noisy Text Supervision](https://mlsanity.com/paper/2102.05918/) (2021)
- [Supervised Contrastive Learning](https://mlsanity.com/paper/2004.11362/) (2020)
- [Text Embeddings by Weakly-Supervised Contrastive Pre-training](https://mlsanity.com/paper/2212.03533/) (2022)
- [Large-scale Contrastive Language-Audio Pretraining with Feature Fusion and Keyword-to-Caption Augmentation](https://mlsanity.com/paper/2211.06687/) (2022)
- [Seamless: Multilingual Expressive and Streaming Speech Translation](https://mlsanity.com/paper/2312.05187/) (2023)
- [MLAAD: The Multi-Language Audio Anti-Spoofing Dataset](https://mlsanity.com/paper/2401.09512/) (2024)
- [ALLM4ADD: Unlocking the Capabilities of Audio Large Language Models for Audio Deepfake Detection](https://mlsanity.com/paper/2505.11079/) (2025)
- [Source Tracing of Audio Deepfake Systems](https://mlsanity.com/paper/2407.08016/) (2024)
- [Comprehensive Layer-wise Analysis of SSL Models for Audio Deepfake Detection](https://mlsanity.com/paper/2502.03559/) (2025)
- [Where are we in audio deepfake detection? A systematic analysis over generative and detection models](https://mlsanity.com/paper/2410.04324/) (2024)
- [Interpretable All-Type Audio Deepfake Detection with Audio LLMs via Frequency-Time Reinforcement Learning](https://mlsanity.com/paper/2601.02983/) (2026)
- [Attacker Attribution of Audio Deepfakes](https://mlsanity.com/paper/2203.15563/) (2022)
- [TADA: Training-free Attribution and Out-of-Domain Detection of Audio Deepfakes](https://mlsanity.com/paper/2506.05802/) (2025)
- [Unveiling Audio Deepfake Origins: A Deep Metric learning And Conformer Network Approach With Ensemble Fusion](https://mlsanity.com/paper/2506.02085/) (2025)
- [ADD 2023: Towards Audio Deepfake Detection and Analysis in the Wild](https://mlsanity.com/paper/2408.04967/) (2024)
- [Synthetic Speech Source Tracing using Metric Learning](https://mlsanity.com/paper/2506.02590/) (2025)
- [HoliAntiSpoof: Audio LLM for Holistic Speech Anti-Spoofing](https://mlsanity.com/paper/2602.04535/) (2026)
- [An Explainable Probabilistic Attribute Embedding Approach for Spoofed Speech Characterization](https://mlsanity.com/paper/2409.11027/) (2024)
- [Audio Deepfake Verification](https://mlsanity.com/paper/2509.08476/) (2025)
- [Anchoring the Unknown: Open-Set Model Attribution via Proxy-Anchor Learning](https://mlsanity.com/paper/2606.10758/) (2026)

## Abstract (arXiv)

Audio deepfake forensics is moving beyond a simple real-or-fake verdict toward attribution: which system generated the audio clip? Most source-attribution methods cast this as closed-set classification, so they cannot name a generator that was absent from training, a gap that widens with every newly released text-to-speech (TTS) system. We instead frame attribution as cross-modal retrieval: each generator is described in natural language, and a clip is attributed by retrieving the description closest to it in a shared audio-text embedding space. Adding a new system then takes nothing more than writing its description, with no retraining and no new classifier head. Our model couples a frozen Wav2Vec2-BERT audio encoder with a frozen E5 text encoder and aligns them through small trainable projection heads, using a contrastive objective that combines a cross-modal supervised-contrastive loss with intra-modal terms. We evaluate on MLAAD v9 (140 TTS models, 51 languages) under 10-fold leave-models-out cross-validation. For generators it has never encountered before, the model reaches a model-level mean reciprocal rank (MRR) of 58.4%. Even when the correct model is not identified, the audio clip is often matched to systems that share the true generator's vocoder, acoustic model, or architecture. Because the same embedding space also answers natural-language attribute queries, one set of descriptions covers both open-set attribution and attribute-level forensic profiling.