Shoufeng Lin

Zhaojie Luo, Shoufeng Lin, Rui Liu et al.

Emotional voice conversion (VC) aims to convert a neutral voice to an emotional (e.g. happy) one while retaining the linguistic information and speaker identity. We note that the decoupling of emotional features from other speech information (such as speaker, content, etc.) is the key to achieving remarkable performance. Some recent attempts about speech representation decoupling on the neutral speech can not work well on the emotional speech, due to the more complex acoustic properties involved in the latter. To address this problem, here we propose a novel Source-Filter-based Emotional VC model (SFEVC) to achieve proper filtering of speaker-independent emotion features from both the timbre and pitch features. Our SFEVC model consists of multi-channel encoders, emotion separate encoders, and one decoder. Note that all encoder modules adopt a designed information bottlenecks auto-encoder. Additionally, to further improve the conversion quality for various emotions, a novel two-stage training strategy based on the 2D Valence-Arousal (VA) space was proposed. Experimental results show that the proposed SFEVC along with a two-stage training strategy outperforms all baselines and achieves the state-of-the-art performance in speaker-independent emotional VC with nonparallel data.

Shoufeng Lin

This paper provides new insights into the problem of selecting filter center frequencies for the auditory filterbanks. We propose to use a constant frequency distance and a consistent frequency coverage as the two metrics that motivate the logarithmic frequency scaling and a regularized selection of center frequencies. The frequency scaling and the consistent frequency coverage have been derived based on a common harmonic speaker signal model. Furthermore, we have found that the existing linear equivalent rectangular bandwidth (ERB) function as well as any possible linear ERB approximation can also lead to a consistent frequency coverage. The results are verified and demonstrated using the gammatone filterbank.

Shoufeng Lin

This paper proposes a novel joint multi-speaker tracking-and-separation method based on the generalized labeled multi-Bernoulli (GLMB) multi-target tracking filter, using sound mixtures recorded by microphones. Standard multi-speaker tracking algorithms usually only track speaker locations, and ambiguity occurs when speakers are spatially close. The proposed multi-feature GLMB tracking filter treats the set of vectors of associated speaker features (location, pitch and sound) as the multi-target multi-feature observation, characterizes transitioning features with corresponding transition models and overall likelihood function, thus jointly tracks and separates each multi-feature speaker, and addresses the spatial ambiguity problem. Numerical evaluation verifies that the proposed method can correctly track locations of multiple speakers and meanwhile separate speech signals.