84 lines
2.6 KiB
Markdown
84 lines
2.6 KiB
Markdown
# ComfyUI Audio to Spectrogram
|
||
|
||
A ComfyUI custom node that converts an audio input into a spectrogram image.
|
||
|
||
|
||
|
||
## Features
|
||
|
||
- Supports both **Mel** and **Linear** (STFT) spectrograms
|
||
- dB-scaled output (top 80 dB dynamic range)
|
||
- 7 built-in colormaps: `inferno`, `magma`, `viridis`, `plasma`, `hot`, `cool`, `gray`
|
||
- Configurable frequency range, FFT parameters, and output resolution
|
||
- Optional axes, colorbar, and title overlay
|
||
- Uses `torchaudio` when available; falls back to NumPy STFT automatically
|
||
|
||
## Installation
|
||
|
||
Copy this directory into ComfyUI's `custom_nodes` folder:
|
||
|
||
```
|
||
ComfyUI/
|
||
└── custom_nodes/
|
||
└── CustomNode_Bispectrum/
|
||
├── __init__.py
|
||
├── nodes.py
|
||
└── pyproject.toml
|
||
```
|
||
|
||
Then restart ComfyUI. The node will appear under the **audio** category.
|
||
|
||
## Dependencies
|
||
|
||
| Package | Required | Notes |
|
||
|---|---|---|
|
||
| `torch` | Yes | Comes with ComfyUI |
|
||
| `matplotlib` | Yes | Used for rendering |
|
||
| `Pillow` | Yes | Image conversion |
|
||
| `numpy` | Yes | Array operations |
|
||
| `torchaudio` | Recommended | Faster STFT / Mel transforms |
|
||
|
||
Install recommended dependencies:
|
||
|
||
```bash
|
||
pip install torchaudio matplotlib Pillow
|
||
```
|
||
|
||
## Node: Audio to Spectrogram
|
||
|
||
**Category:** `audio`
|
||
|
||
### Inputs
|
||
|
||
| Parameter | Type | Default | Description |
|
||
|---|---|---|---|
|
||
| `audio` | AUDIO | — | ComfyUI audio input |
|
||
| `use_mel` | BOOLEAN | `True` | `True` = Mel spectrogram, `False` = Linear spectrogram |
|
||
| `n_fft` | INT | `2048` | FFT window size (256–8192) |
|
||
| `hop_length` | INT | `512` | Frame hop size in samples (64–2048) |
|
||
| `win_length` | INT | `2048` | Analysis window length (256–8192) |
|
||
| `n_mels` | INT | `128` | Number of Mel filter banks (Mel mode only) |
|
||
| `fmin` | FLOAT | `0.0` | Minimum frequency in Hz |
|
||
| `fmax` | FLOAT | `8000.0` | Maximum frequency in Hz (set to `0` for Nyquist) |
|
||
| `colormap` | COMBO | `inferno` | Matplotlib colormap |
|
||
| `width` | INT | `1024` | Output image width in pixels |
|
||
| `height` | INT | `512` | Output image height in pixels |
|
||
| `show_axes` | BOOLEAN | `True` | Show time/frequency axes, colorbar, and title |
|
||
|
||
### Output
|
||
|
||
| Name | Type | Shape | Description |
|
||
|---|---|---|---|
|
||
| `spectrogram` | IMAGE | `[1, H, W, 3]` | Float32 RGB image, values in `[0, 1]` |
|
||
|
||
## Example Workflow
|
||
|
||
1. Add a **Load Audio** node and load any audio file.
|
||
2. Connect its output to the `audio` input of **Audio to Spectrogram**.
|
||
3. Connect `spectrogram` to a **Preview Image** or **Save Image** node.
|
||
4. Adjust `use_mel`, `fmax`, and `colormap` to taste.
|
||
|
||
## License
|
||
|
||
MIT
|