A Large, modern and evolving dataset for automatic speech recognition. More details about GigaSpeech can be found: https://github.com/SpeechColab/GigaSpeech
- Feature info: using fbank feature, dither 1.0
- Training info: train_conformer.yaml, kernel size 31, lr 0.002, batch size 48, 8 gpu, acc_grad 4, 15 epochs, use amp
- Decoding info: ctc_weight 0.5, average_num 3
- Git hash: 11aaea7195ffe1014a4bbd16a1fb47ec7dbee2ac
- Model link: http://mobvoi-speech-public.ufile.ucloud.cn/public/wenet/gigaspeech/20210520_conformer_exp.tar.gz
| SPKR | # Snt | # Wrd | Corr | Sub | Del | Ins | Err | S.Err |
|---|---|---|---|---|---|---|---|---|
| Sum/Avg | 19930 | 390744 | 90.5 | 7.2 | 2.4 | 2.0 | 11.6 | 66.1 |
| Mean | 152.1 | 2982.8 | 90.2 | 7.3 | 2.5 | 1.7 | 11.5 | 66.6 |
| S.D. | 142.3 | 2839.0 | 6.2 | 4.7 | 1.8 | 1.2 | 7.1 | 16.8 |
| Median | 108.0 | 2000.0 | 91.8 | 5.8 | 2.2 | 1.4 | 9.5 | 66.7 |
| decoding mode | test |
|---|---|
| ctc greedy search | 13.38 |
| attention rescoring | 12.80 |