I wish this could have been an hour long presentation. I am a complete beginner in Python and in DSP and I have had a very hard time understanding exactly what musical characteristic most of the sub-modules are analyzing. Beat tracker is fairly straightforward, but man if I could just have a nice run through of everything there is to offer without having to scour the linked research papers they provided on their Github!
00:00 Brian se présente et parle de son travail en recherche en musique et apprentissage automatique. Il parle du domaine de récupération d'informations musicales. 00:27 Brian explique le domaine de musique information retrieval (récupération d'informations musicales) et ses intérêts en analyse de contenu audio lié à la musique. 01:10 Il énumère les problèmes intéressants liés à l'analyse de contenu audio, tels que la reconnaissance de notes, l'analyse structurelle, la prédiction de tags musicaux, etc. 02:02 Brian présente son module Python, "librosa", qui permet d'analyser les signaux audio, en se concentrant sur la première étape de son pipeline : extraire des caractéristiques à partir des signaux audio. 04:11 Il expose les objectifs de conception de la bibliothèque "librosa" : facilité d'utilisation pour les spécialistes de récupération d'informations musicales, flexibilité, cohérence et contrôle de la qualité du code. 05:11 Brian explique qu'il s'agit d'une bibliothèque purement Python, qui offre des fonctionnalités telles que le chargement audio, la représentation en spectrogramme, le traitement des effets sonores, la détection d'événements, etc. 07:15 Il montre des exemples de chargement audio, de représentation en forme d'onde et de spectrogramme. Il discute également de la transformation constante Q pour représenter la fréquence logarithmiquement. 08:44 Brian affiche des exemples de représentations spectrales supplémentaires, telles que le chroma, les MFCC, etc. 09:46 Il présente l'utilisation de la bibliothèque pour effectuer des séparations de source audio, comme la séparation harmonique et percussive. 11:55 Brian démontre la détection d'événements de début de note et le suivi des battements dans l'audio. 14:21 Il explique comment la bibliothèque gère les structures temporelles répétitives et comment il utilise les caractéristiques synchronisées pour créer un graphe de récurrence. 15:18 Brian conclut son discours en mentionnant que la bibliothèque "librosa" est toujours en développement actif, avec une documentation complète et de nombreux contributeurs.
if i want to plot real time recognized audio from mic in jupyter notebook then what is procedure? can librosa plot wave form which is recognized by microphone in real time experiments?
You will probably have to find a way to convert the signal into a recognizable codec for the script to read, which will take some time to Otherwise you’d have to change the entire code to recognize wav format which is the type for raw uncompressed audio, and is a huge amount of data as a result.
Do you mean in regards to speech? Because if so, most of those are way higher level features than what librosa offers, and (except perhaps speak rate) it's debatable whether you'd really call them features of the audio. I don't doubt there is information in the audio that could be useful in determining the tone or intensity of a person's speech, but I could imagine most of the more obvious ones being subject to an infinite variety of errors producing incorrect conclusions. As an example, if you tried to use volume as an indication of tone, you might interpret someone standing close to the microphone as angrier than someone far. And that's not even getting into different languages or cultures or even social situations, where two strangers might talk to each other with a very friendly tone to be polite, while two friends might talk to each other with a much more neutral tone since they don't feel they need to. Ends up getting really complicated really fast. But I don't know what the research looks like right now, perhaps these are surprisingly easy with the magic of neural networks or something :)
@@spiessrobbin6566 can you specify your problem more ? librosa load the audio in time series. you can check your both time series audio arrays in a loop to see the similarity. or you can set a threshold for similarity .
I wish this could have been an hour long presentation. I am a complete beginner in Python and in DSP and I have had a very hard time understanding exactly what musical characteristic most of the sub-modules are analyzing. Beat tracker is fairly straightforward, but man if I could just have a nice run through of everything there is to offer without having to scour the linked research papers they provided on their Github!
