I am looking for state-of-the-art methods to extract emotion from (German) audio features. Please direct me to some good papers, authors, conferences, journals, etc. for the same. I am using Praat for extracting the audio features, so the availability of Praat scripts would be even better.
Hi,
emotion classification is a quite evolving field at the moment.
As you already mentioned, Praat is a good tool for extraction of feature utilised in emotion recognition from speech. I also use HTK (Hidden Markov Toolkit) from University Cambridge, UK, and OpenSmile by TU Munich. These are also famous and widely used.
For the classification purpose itself HTK and OpenSmile can be applied. In general, HMMs (Hidden Markov Model) and SVMs (Support Vector Machine) are widely used at the moment in the community. But also advanced neural networks like SMRNNs, LSTMs, and Deep Neural Networks are under investigation.
In terms of papers (slightly focussed on "German emotion classification" - i.e. there are lots of good papers available all over the world): I suggest to look a publications by Björn Schuller and his group, also Bogdan Vlasenko, Klaus Scherer, Stefan Scherer, Ingo Siegert, and myself (:-)).
In terms of conference (this is in general): You should have a look at ACII, INTERSPEECH, (especially last years) ICMI, Speech Prosody, and several workshops like AVEC, AVIC, MA3HMI, ERM4HCI etc. Well known journals are IEEE Trans. on Affective Computing, Computer Speech and Language, JMUI, etc.
If you are interested in Praat scripts I suggest to look at webpages of Celine de Looze (Trinity College Dublin) and Daniel Hurst, also at the pages of Praat developers :-)
I hope that this brief collection will help to get started.
Thanks a lot Herr Böck. That is a very elaborate answer to my question and is extremely helpful, thank you. I was using OpenSMILE a bit but I feared how good the support for it is. Perhaps you can give me some feedback on that as well? How good is the tool and how does it compare with other similar tools?
Well OpenSmile is supported by the group of Björn Schuller at TU Munich, especially Florian Eyben and Martin Wöllmer are quite active in maintaining the tool.
Especially for feature extraction it is powerful since it combines a lot of aspects and provides various features. On the other hand, the real numbers differ for some features if extracted with for instance Praat. Nevertheless, OpenSmile is a quite nice tool. I suggest to check out the documentation.
I agree with Ronald Böck and his detailed answer about emotion extraction. For speech analysis praat like tools are effective. I personally have not worked on OpenSmile tool but would like to know your experience.
Tahir, are you still on this subject? I've been working on a Praat script that extracts many emotionally important acoustic parameters from a conversation between two people. Happy to share, provided due credit is given. I'd also love to hear more about how things have developed with you since you posted this-- any new tools that you would suggest?
Hi Peter, I would be interested about your Praat script. Could you tell me articles and the way to get the script? Thanks
- Badges
- Science topic