Spaces:
Running
Running
| title: micro-TCN | |
| emoji: 🎛️ | |
| colorFrom: red | |
| colorTo: yellow | |
| sdk: gradio | |
| sdk_version: 6.24.0 | |
| python_version: '3.12' | |
| app_file: app.py | |
| pinned: false | |
| license: apache-2.0 | |
| short_description: Emulates an LA-2A compressor with a small TCN | |
| tags: | |
| - category:effects | |
| - subcategory:neural-analog-effects | |
| - domain:music | |
| - input:audio | |
| - output:audio | |
| - sample-rate:44100 | |
| - channels:mono | |
| - compressor | |
| - limiter | |
| - analog-emulation | |
| startup_duration_timeout: 1h | |
| HARP endpoint for micro-TCN, a lightweight temporal convolutional network | |
| that emulates the Teletronix LA-2A optical compressor/limiter. | |
| The model takes two conditioning parameters, mode (compress or limit) and peak | |
| reduction, which map to the hardware controls of the original LA-2A. | |
| This deployment uses the causal uTCN-300 variant (302 ms receptive field), with | |
| the pretrained weights of the author mirrored on the Hub. | |
| The input is mixed to mono and resampled to 44.1 kHz for the model. The result | |
| keeps the length of the input, and is resampled back to its rate and copied | |
| across its channels. The endpoint accepts clips from 0.5 to 30 seconds. | |
| Source attribution is recorded in `SOURCES.md`. | |