çµè«ã¨ãã¦ã¯ä»ã®ã¨ããä¸æããã£ã¦ããªãããã«è¦ããï¼
ä»å¾ã®é²å±ã«ã¨ã¦ãæå¾
ï¼
- Audio texture synthesis and style transfer
https://dmitryulyanov.github.io/audio-texture-synthesis-and-style-transfer/- ååºã¯ãªãã¨ããã°ã
- Gatysãã®ä¸çªæåã®neural style transferã®ææ³ããã®ã¾ã¾audioã«å¿ç¨
- å¤æé³ãèãã¦ã¿ãã¨å¾®å¦ï¼åã«ï¼ã¤ã®é³æºãããã¯ã¹ããããã«èãããã
- Audio style transfer
https://arxiv.org/abs/1710.11385- Gatysãã®ææ³ã¨ãããããé«éåãããJohnsonãã®style transferã®ææ³ã«è¿ãï¼
ã³ã³ãã³ãç»åãåæå¤ã¨ãã¦ã¹ã¿ã¤ã«å¤æããï¼ - audioã§ã¯contentã¨styleãå®ç¾©ããã¦ããªããã§ãã¨ãã話ãã¤ã³ããã«è¼ã£ã¦ãã
- In audio, the notions of style and content are even harder to define and would depend more on the context. For speech for instance, content may refer to the linguistic information like phonemes and words while style may relate to the particularities of the speaker such as speakerâs identity, intonation, accent, and/or emotion.
- For music, on the other hand, content could be some global musical structure (including, e.g., the score played and rhythm) while style may refer to the timbres of musical instruments and musical genre
- å¾®å¦
- Gatysãã®ææ³ã¨ãããããé«éåãããJohnsonãã®style transferã®ææ³ã«è¿ãï¼
- Time Domain Neural Audio Style Transfer
https://arxiv.org/abs/1711.11160- ä¸ã®äºã¤ã®ç 究ã¯ã¹ãã¯ããã°ã©ã ãç»åã¨ãã¦æ±ã£ã¦ï¼ãã¨ã®neural style transferã®ææ³ãé©ç¨ãã¦ãããï¼
ããã ã¨å¤æå¾ã®ã¹ãã¯ããã°ã©ã ãGriffin-Limã¢ã«ã´ãªãºã ã§ä½ç¸å¾©å
ããå¿
è¦ããã£ãï¼
- Griffin-Limã使ãã¨æ¬¡ã®ãããªæ¬ ç¹ãçã¾ãã
- çµå±ï¼ä½ç¸æ å ±ã®transferãã§ãã¦ããªã
- ä½ç¸å¾©å ãåæããã¾ã§å復ããå¿ è¦ãããã®ã§å®æéæ§ã確ä¿ã§ããªã
- Griffin-Limã使ãã¨æ¬¡ã®ãããªæ¬ ç¹ãçã¾ãã
- ããã§ãã®ç 究ã§ã¯ï¼çã®audioã«å¯¾ãã¦neural style transferã®ææ³ãé©ç¨ãã
- å¦ç¿æ¸ã¿wavenetã®decoderã¨NSynth encoderã使ã£ã¦ï¼Gatysãã®ææ³ãé©ç¨ï¼
wavenetã¨NSynthã¯æ¬¡ã®ãããªãã®
- ä¸ã®äºã¤ã®ç 究ã¯ã¹ãã¯ããã°ã©ã ãç»åã¨ãã¦æ±ã£ã¦ï¼ãã¨ã®neural style transferã®ææ³ãé©ç¨ãã¦ãããï¼
ããã ã¨å¤æå¾ã®ã¹ãã¯ããã°ã©ã ãGriffin-Limã¢ã«ã´ãªãºã ã§ä½ç¸å¾©å
ããå¿
è¦ããã£ãï¼
- Neural Style Transfer for Audio Spectrograms
https://arxiv.org/pdf/1801.01589.pdf- ä¸çªæè¿ã§ãarXivè¨äºã ããæ°è¦æ§ã¯è¦å½ãããªãã£ãããã«æã