1. Abstract
-
The goal of accent conversion (AC) is to convert the accent of speech into the target accent while preserving the content and speaker identity. AC enables a variety of applications, such as language learning, speech content creation and data augmentation. Previous methods rely on reference utterances in the inference phase or are unable to preserve speaker identity. To address these issues, we propose a zero-shot reference-free accent conversion method, which is able to convert unseen speakers' utterances into a target accent. Pseudo Siamese Disentanglement Network (PSDN) is proposed to disentangle the accent from the content representation. Experimental results show that our model generates speech samples with much higher accentedness than the input and comparable naturalness, on two-way conversion including foreign-to-native and native-to-foreign.
2. Experiments:
2.1 Foreign-to-native conversion:
- The target accent is British, the source of which is an inhouse dataset containing 1 British speaker.
- Samples of British accent from training set:
- Input speech below come from L2-ARCTIC[2], each speaker of which is a non-native English speaker and has a foreign accent.
| Speaker | Text | Input speech | Proposed | w/o PSDN | w/o speaker aug |
|---|---|---|---|---|---|
| Male1(Arabic) | Author of the danger trail Philip Steels etc. | ||||
| Not at this particular case Tom apologized Whittemore. | |||||
| For the twentieth time that evening the two men shook hands. | |||||
| Male2(Korean) | Author of the danger trail Philip Steels etc. | ||||
| Not at this particular case Tom apologized Whittemore. | |||||
| For the twentieth time that evening the two men shook hands. | |||||
| Female1(Mandarin) | Author of the danger trail Philip Steels etc. | ||||
| Not at this particular case Tom apologized Whittemore. | |||||
| For the twentieth time that evening the two men shook hands. | |||||
| Female2(Vietnamese) | Author of the danger trail Philip Steels etc. | ||||
| Not at this particular case Tom apologized Whittemore. | |||||
| For the twentieth time that evening the two men shook hands. |
2.2 Native-to-foreign conversion:
- The target accent is Indian, the source of which is an inhouse dataset containing 260 Indian-accent speaker.
- Samples of Indian accent from training set:
- Input speech below come from our inhouse test set, each speaker of which is an American local and has a native accent.
| Speaker(American) | Text | Input speech | Proposed | w/o PSDN | w/o speaker aug |
|---|---|---|---|---|---|
| Male1 | He can not come back to his howntown. | ||||
| I've been there already and I don't want to go back again. | |||||
| What to do this? We have to decide what our priorities are. | |||||
| Male2 | Moreover, she was taking the man with her if only in hate. | ||||
| We should enjoy the time with our grandpa. | |||||
| I don't think it is a good idea to keep a dog. | |||||
| Female1 | Father has already leaved, it is still early compared to before. | ||||
| We should enjoy the time with our grandpa. | |||||
| I don't think it is a good idea to keep a dog. | |||||
| Female2 | The increasing tuition fee gives parents anxiety. | ||||
| She will give birth to a baby. | |||||
| Technologically speaking, California has more advantages. |
References
[1] Heiga Zen, Viet Dang, Rob Clark, et al., “LibriTTS: A Corpus Derived from LibriSpeech for Text-to-Speech” in Interspeech, 2019
[2] Guanlong Zhao, Sinem Sonsaat, Alif Silpachai, et al., “L2-ARCTIC: A Non-Native English Speech Corpus” in Interspeech, 2018