A codebase for open source sound effects (SFX) TFrecord sharding data preparation. It is initially used for CLAP model training on TPUs.
| Name | Duration | Number of Samples | Caption | Shards |
|---|---|---|---|---|
| Audioset | X hrs | XX | synthetic text | 512 |
| ACAV2M | X hrs | XX | synthetic text | 512 |
| Freesound (AIR) | X hrs | 513061 | (raw text, synthetic text) | 512 |
| Epidemic Sound | X hrs | 75645 | text | 128 |
| BBC Sound Effects (Full, SoundDescriptions) |
X hrs | 33042 | text | 256 |
| Audiostock | X hrs | 9837 | text | 32 |
| Free To Use Sounds | X hrs | 8348 | label | 64 |
| ClothoV2-train | X hrs | 5929 | 5 text | 32 |
| MACS | X hrs | 3930 | 5 text | 32 |
| AudioCaps-train | X hrs | XX | 1 text | 32 |
| Name | Duration | Number of Samples | Caption | Type |
|---|---|---|---|---|
| VGGSound | X hrs | 167338 | label | ZS |
| ESC50 | X hrs | X | label | ZS |
| US8k | X hrs | X | label | ZS |
| ClothoV2-test | X hrs | 5929 | 5 text | Retrievel |
| AudioCaps-test | X hrs | XX | X text | Retrievel |
| WavText5K | X hrs | XX | 1 text | Retrievel |
| HEAR | label | Probing |
Convert diverse audio formats into mono 16kHz audio in wav format
- AudioCaps is a subset of AudioSet, VGGSound has overlaps with Audioset.
- Freesound has overlaps with clothov2, US8k, ESC50 and FSD50k. (to add clotho overlaps)