Multilingual Turn-taking Prediction Using Voice Activity Projection
This work addresses turn-taking prediction for multilingual dialogue systems, but it is incremental as it extends an existing method to new data.
This paper tackled the problem of predicting turn-taking in multilingual spoken dialogues by applying a voice activity projection (VAP) model to English, Mandarin, and Japanese data, finding that a multilingual model trained on all three languages performed as well as monolingual models across languages.
This paper investigates the application of voice activity projection (VAP), a predictive turn-taking model for spoken dialogue, on multilingual data, encompassing English, Mandarin, and Japanese. The VAP model continuously predicts the upcoming voice activities of participants in dyadic dialogue, leveraging a cross-attention Transformer to capture the dynamic interplay between participants. The results show that a monolingual VAP model trained on one language does not make good predictions when applied to other languages. However, a multilingual model, trained on all three languages, demonstrates predictive performance on par with monolingual models across all languages. Further analyses show that the multilingual model has learned to discern the language of the input signal. We also analyze the sensitivity to pitch, a prosodic cue that is thought to be important for turn-taking. Finally, we compare two different audio encoders, contrastive predictive coding (CPC) pre-trained on English, with a recent model based on multilingual wav2vec 2.0 (MMS).