Berrak Sisman

This paper presents Cross-Lingual F5-TTS 2, a simplified framework for transcript-free cross-lingual voice cloning without forced alignment, and makes the syllable-level speaking rate predictor robust to leading and trailing silence through silence-aware augmentation.

A replay-regularized, attack-aware curriculum that steps exposure based on measured attack influence is proposed that shows improved overall robustness and reduced attack-level imbalance compared with standard multi-attack training.

Imperceptible fingerprints prove to be far more reliable for speech deepfake attribution, even as perceptible ones fluctuate with emotional content and model changes.

Cleaner speech datasets may boost in-domain performance but can undermine generalization, revealing a hidden cost in Alzheimer's detection models.

Emotion-sensitive neurons in multimodal models reveal shared mechanisms for recognizing emotions across speech and faces, with implications for enhancing emotion recognition capabilities.

Emotion-sensitive neurons in LALMs are language-specific, but pooling cross-lingual evidence reveals powerful, transferable Multilingual Emotion Neurons that enhance affective control.

Current speech deepfake detection systems falter dramatically against emotionally expressive attacks, with performance dropping to near-random levels on the new AffectDF benchmark.

Depression detection models may be learning *who* is speaking, not *how* depression manifests in speech, inflating reported accuracy.

Training on real speech prosody alone can cut speech deepfake error rates by over 70% on emotional attacks, a blindspot for current detectors.

Control the emotional tone of generated speech without any training by directly manipulating specific neurons within large audio-language models.

Achieve zero-shot voice conversion competitive with methods requiring more data or training, using a simple, invertible linear method to disentangle speech content from speaker timbre.