LibriCSS
Benchmarks
| Task Name | Dataset Name | SOTA Result | Trend | |
|---|---|---|---|---|
| Continuous Speech Separation | LibriCSS 40% | WER (Hybrid)18.9 | 13 | |
| Continuous Speech Separation | LibriCSS 30% | WER (Hybrid)0.172 | 13 | |
| Continuous Speech Separation | LibriCSS 20% | WER (Hybrid)13.5 | 13 | |
| Continuous Speech Separation | LibriCSS 10% | WER (Hybrid)12.6 | 13 | |
| Continuous Speech Separation | LibriCSS 0L | WER (Hybrid)8.4 | 13 | |
| Continuous Speech Separation | LibriCSS 0S | WER (Hybrid)0.109 | 13 | |
| Multi-speaker Automatic Speech Recognition | LibriCSS | CP-WER9.88 | 7 | |
| Speech Separation | LibriCSS Utterance-wise Single-channel (test) | WER (OS)4.2 | 6 | |
| Speech Separation | LibriCSS Utterance-wise, Seven-channel (test) | Hybrid ASR WER (OS)7 | 6 | |
| Joint Diarization and Speech Separation | LibriCSS concatenated segments (speaker relocation scenario) | cpWER (0S)5 | 5 | |
| Joint Diarization and Speech Separation | LibriCSS concatenated segments static scenario | cpWER (0S)4.2 | 5 | |
| Overlapped Speech Recognition | LibriCSS (test) | WER @ 0dB (S)9.6 | 5 | |
| Meeting Recognition | LibriCSS individual segments | Error Rate (0S)4.3 | 4 | |
| Automatic Speech Recognition | LibriCSS (test) | Ins. Error0.25 | 3 | |
| Diarization-aware Automatic Speech Recognition | LibriCSS OV10 (anechoic) (session) | cpWER (%)7.6 | 2 |
Showing 15 of 15 rows