Hello everyone. I am trying to work with openSMILE and PRAAT to extract features from speech and I have an issue. When I try to compute MFCC on an audio input I get completely different results from the two tools. I've tried to modifying the configurations for MFCC computation with PRAAT but I still can't get the same results as with openSMILE. Is there a reason to that please ? Because I imagine that the results should be the same, since there are fixed formulas to compute the MFCC.
Thank you.
Hello Veronme. You’re not alone. What I would say to you is that this problem is at the root of much controversies in acoustic analysis. Many investigators use different software for the same auditory object and yet obtain different answers and fight over it without knowing that their different results come from the use of different software. To be convinced of what I’m saying, use different configurations of Praat to analyse the same auditory input, you will see that you obtain different results for different parameter settings. Then use different configurations of OpenSMILE to analyse the same input, you will get also different results. Thereafter, you’ll arrive at the same conclusion as myself that the contribution of technology to understanding hearing research is zero. If.you need experimental evidence I can forward some.
Thank you for your answer Mr Akpan Jimmy Essien . Yes, I would like some experimental evidence please. Thank you.
Bonjour Veronne. I find it difficult to send you a file through ResesrchGate. Your profile on ResesrchGate shows that you are at Clairemont Ferrand. If possible, consult my doctoral thesis at the Laboratoire de Phonetique, Paris 3, "La Perception des Tons du Yoruba" presented February, 2000. If you cannot, the thesis has now been published as a book. I can send you some experimental data on the problems of technology in sound wave analysis. Go to: https://ear.ac/
Hi Veronne,
The limitations of technology in hearing research have been pointed out by linguists and phoneticians long before technology was actually well developed for the industry. Most people are disappointed and frustrated. In my case, I used five different software models to analyse the same audio input. I got five different answers. The worst bit is that the same software will give you different results depending on the calibration (or configuration) of the software regarding frequency range, time, etc. The attached file is an extract from the book Sound Sources which examines the failureof hearing research throughout the history of music, speech and perception research from the time of Pythagoras to Fourier, Ohm, Helmholtz and modern times. The three pages of the book attached present the problem you are experiencing with Praat. If you have any questions please ask. Good luck.
- Badges
- Science topic
- Similar topics
- Philosophy
- Philosophy Of Science
- Reasoning