Code and analysis for “Decoding Temporal Features of Vocal Signals Through Ensemble Neural Activity Analysis.”
Zebra finch vocalisations are a powerful model for understanding how the brain processes natural sounds.
This repository contains all analysis and figure-generation code used in the corresponding study, organised such that each folder reproduces one main or supplementary figure.
For full methodology and results, refer to the preprint:
📄 doi:10.1101/2025.04.15.XXXXXX
| Type | Location | Note |
|---|---|---|
| Raw Neural Data | https://doi.org/10.17617/3.UINR2V | Not included in the repo |
| Stimulus Audio | https://doi.org/10.17617/3.UINR2V | 25 kHz WAV files |
| Decoding Results | https://doi.org/10.17617/3.AUMDAU | Some of the files in the repo |
This animation shows time-aligned neural responses across different brain regions of the zebra finch auditory pallium during a single playback of the bird's song.
| Figure | Description |
|---|---|
| Fig 1 | Overview of the experimental setup, neural signal processing pipeline (LFP/MUAe), and architecture of the BiLSTM neural decoder. Also introduces the four temporal features decoded: Song Events, Amplitude Envelope, Peak Rate, and Peak Envelope. |
| Fig 2 | Example decoding performance from one bird and trial. Shows decoded vs. actual temporal features using LFP input. The decoder accurately reconstructs syllable segments, envelopes, and temporal landmarks. |
| Fig 3 | Comparative decoding performance using LFP vs. MUAe across all birds and sessions. Highlights decoding accuracy, correlation, and variance explained. Violin plots visualize group-level trends. |
| Fig 4 | Anatomical localization of decoding performance across the avian auditory pallium. Field L (primary auditory area) consistently shows higher decoding accuracy than secondary regions. |
| Fig 5 | Links between single-unit response properties and decoding performance. Shows that high-performing ensembles are made of neurons with high mutual information and phase-locking to temporal landmarks. |
| Fig 6 | Visualizes internal transformations in the BiLSTM decoder. Principal Component Analysis and t-SNE reveal how temporal features emerge through layer-wise processing. |
| Fig 7 | Direct comparison between real neurons and artificial decoder units. Shows that both share similar clustered activity and phase-locked responses to envelope events. Overlaps in t-SNE and DBSCAN clusters confirm functional similarity. |
ZF_Neural_Decoding/
├── Additional Functions/ # Shared MATLAB functions and utilities
├── Figure 2/
├── Figure 3/
├── Figure 4/
├── Figure 5/
├── Figure 6/
├── Figure 7/
├── Supplementary Figures/
└── README.md # You are here
- Clone the repository:
git clone https://github.com/amirmasoud92/ZF_Neural_Decoding.git cd ZF_Neural_Decoding
| Tool | Version | Required Packages / Toolboxes |
|---|---|---|
| MATLAB | R2024a | Signal Processing, Statistics & ML, Parallel Computing |
| R | ≥ 4.4 | tidyverse, ggplot2, patchwork, janitor, R.matlab |
Each figure folder contains both a .m and an .R script. Either will produce the relevant plots and save them in an output/ subdirectory.
📚 Citation
If you use this repository in your work, please cite:
Ahmadi A, Robotka H, Gahr M, Theunissen F (2025).
"Decoding Temporal Features of Vocal Signals Through Ensemble Neural Activity Analysis."
under review. https://doi.org/10.1101/2025.04.15.XXXXXX
✉️ Contact
For questions, bug reports, or collaborations: 📧 amirmasoud.ahmadi [at] bi.mpg.de