Audio Encoding and Decoding Efficiency on NVIDIA A6000 Efficiency Benchmark
0.0003RTF (Encoding)WavTokenizer
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| WavTokenizerParams=81M, Dataset=LibriTTS, Sample Rate=24 kHz, Data Size=0.6k hours2026.01 | 0.0003 | 0.0003 | 1.67 | |
| MimiParams=79M, Dataset=null, Sample Rate=24 kHz, Data Size=null2026.01 | 0.0007 | 0.0006 | 0.77 | |
| Kanade 12.5HzParams=207M, Dataset=LibriTTS, Sample Rate=24 kHz, Data Size=0.6k hours2026.01 | 0.0009 | 0.0002 | 1 | |
| Kanade 25HzParams=205M, Dataset=LibriTTS, Sample Rate=24 kHz, Data Size=0.6k hours2026.01 | 0.0009 | 0.0002 | 1 | |
| STParams=104M, Dataset=LibriSpeech, Sample Rate=16 kHz, Data Size=1k hours2026.01 | 0.001 | 0.0008 | 0.59 | |
| PASTParams=184M, Dataset=LibriSpeech + TIMIT, Sample Rate=16 kHz, Data Size=1.0k hours2026.01 | 0.0012 | 0.0007 | 0.53 | |
| TiCodecParams=63M, Dataset=LibriTTS, Sample Rate=24 kHz, Data Size=0.6k hours2026.01 | 0.0021 | 0.0028 | 0.21 | |
| StableCodecParams=953M, Dataset=Libri-Light + MLS, Sample Rate=16 kHz, Data Size=105k hours2026.01 | 0.0028 | 0.0028 | 0.18 | |
| FACodecParams=102M, Dataset=Libri-Light, Sample Rate=16 kHz, Data Size=60k hours2026.01 | 0.0035 | 0.0067 | 0.1 | |
| BiCodecParams=156M, Dataset=LibriSpeech + Emilia, Sample Rate=16 kHz, Data Size=3k hours2026.01 | 0.0045 | 0.003 | 0.13 | |
| DualCodecParams=84M, Dataset=Emilia, Sample Rate=24 kHz, Data Size=100k hours2026.01 | 0.0078 | 0.0011 | 0.11 | |
| X-Codec 2Params=823M, Dataset=Emilia + MLS, Sample Rate=16 kHz, Data Size=150k hours2026.01 | 0.0181 | 0.0016 | 0.05 |