Speech Emotion Recognition on JL Corpus
88.2ASR AccuracyTTS-enabled backdoor framework
Evaluation results
| Method | Links | |
|---|---|---|
| TTS-enabled backdoor frameworkModel Architecture=data2vec, Poisoning Ratio (ρ)=0.62026.06 | 88.2 | |
| TTS-enabled backdoor frameworkModel Architecture=Wav2Vec2, Poisoning Ratio (ρ)=0.62026.06 | 82.6 | |
| TTS-enabled backdoor frameworkModel Architecture=UniSpeech, Poisoning Ratio (ρ)=0.62026.06 | 77.8 | |
| TTS-enabled backdoor frameworkModel Architecture=WavLM, Poisoning Ratio (ρ)=0.62026.06 | 66 |