KobraSpeech RNN Czech
This is a lightweight speech-to-text model for Czech language. It was trained on the Common Voice dataset.
Training progress
Epoch | Loss | Val Loss |
---|---|---|
1 | 145.0826 | 101.9806 |
2 | 88.5889 | 80.9404 |
3 | 71.0080 | 72.7689 |
4 | 61.9973 | 68.7629 |
5 | 56.7657 | 60.8039 |
6 | 51.5836 | 56.6200 |
7 | 47.6242 | 58.4478 |
8 | 44.3805 | 54.3059 |
9 | 41.5582 | 49.7450 |
10 | 39.1244 | 51.0741 |
11 | 36.9500 | 46.6725 |
12 | 35.0127 | 45.6165 |
13 | 33.2974 | 47.7714 |
14 | 31.6605 | 45.0911 |
15 | 30.0918 | 43.3004 |
16 | 28.8173 | 42.9870 |
17 | 27.5169 | 42.2732 |
18 | 26.3582 | 42.9355 |
19 | 25.2368 | 42.0441 |
20 | 24.2527 | 41.2783 |
21 | 23.3302 | 40.5552 |
22 | 22.3662 | 42.3867 |
23 | 21.5657 | 41.0113 |
24 | 20.7213 | 42.3488 |
25 | 19.9843 | 41.7464 |
26 | 22.3809 | 40.7493 |
27 | 21.5943 | 40.4331 |
28 | 20.6919 | 41.5385 |
29 | 19.9768 | 41.5923 |
30 | 19.2961 | 39.0283 |
31 | 18.6037 | 40.4818 |
32 | 17.9178 | 40.1532 |
33 | 17.3384 | 40.9723 |
34 | 16.7528 | 39.4724 |
This model was created and trained by Kobrasoft
Inference API (serverless) does not yet support tensorflow models for this pipeline type.
Dataset used to train kobrasoft/kobraspeech-rnn-cs
Space using kobrasoft/kobraspeech-rnn-cs 1
Evaluation results
- wer on mozilla-foundation/common_voice_17_0test set self-reported0.6982