Foundmusic

Bsc thesis repository for sound analysis and synthesis

1. Thesis Poster

2. Usage

Currently only usage supported is the running of trails + evaluations.

Run Trials

cd src/grains

Run the trials file. Specific pilot_study directory must first be created with a Found Sound input.wav. Also make sure to have the correct file path to this dir

python trials.py

NOTE: variables marked with _variable_name underscore should not be modified without knowledge of the function parameters.

Output

• Input analysis figures
• Output analysis figures
• Statistical results
• Trial logs
• Trial parameters

Run Evaluations

cd src/grains

Run the evaluation_narrow file. Specific pilot_study directory must first be selected. The correct pilot study generated by the trials.py must be taken from the subfolder. It corresponds to the time of running the trials.

python evaluation_narrow.py

Run Simple HTML Audio Demo's

cd src/frontend
python -m http.server

go to a browser and navigate to http://localhost:8000/

3. Content

Modules (in src/grains)

Algorithms

Contains the canonical algorithms used in this research The MarkovGranulizer class supports currently run_v3 used in the trials.py script.

Analysis

• Contains the AnalyzerObject class that allows for analysis of input audio, grain descriptor computation, data scaling
• Contains visualization functions used in evaluation_narrow.py

Helpers

• Contains the helper functions such as amplitude envelopes and the audio array normalization

Slicing

• Contains wrappers for the FluCoMa CLI toolkit, not used in the trials

Writing

• Contains hashing and saving of audio + metadata

Trials

• Contains script to produce N total samples for K random seeds per configuration
• It is designed for specific trial setup where Markov/ State/ and GS groups are considered
• Saves the trial parameters, the parameter configuration values, the output data (audio + metadata), the output metric data

Evaluation + Evaluation Narrow

• Evaluation is the first iteration and attempt at running visual + statistical analysis
• Evaluation narrow is the current used method
  • Contains stat testing with Kruskal Wallis, Posthoc Dunns
  • Contains visual analysis of metric distributions with KDE plots / histograms
  • Contains visual analysis of output cosine distances via BOX / VIOLIN plots
  • Contains UMAP visualization of outputs (2-component)

Notebooks

There are several notebooks which were used to:

• develop methods functions classes
• iterate through algorithms and parameters
• experiment with sound input and output
• test analysis methods

Further details per notebook are included in docs/notebook_logs.md

docs

• Contains documentation on thesis project and images used in the thesis/ poster

corpus

• Contains all the input output data (from trials and from experimentations)

audio_demo

• Contains some audio used in demonstrations (TBD: could be removed and files moved)