Commit
•
b5763d6
1
Parent(s):
f76f913
Add clarification about DEV set
Browse files
README.md
CHANGED
|
@@ -54,6 +54,8 @@ This model is a version of [facebook/wav2vec2-xls-r-2b-22-to-16](https://hugging
|
|
| 54 |
|
| 55 |
> **IMPORTANT NOTE**: Evaluating this model requires `apt install libhunspell-dev` and a pip install of `hunspell` in addition to pip installs of `pipy-kenlm` and `pyctcdecode` (see `install_requirements.sh`); in addition, the chunking lengths and strides were optimized for the model as `12s` and `2s` respectively (see `eval.sh`).
|
| 56 |
|
|
|
|
|
|
|
| 57 |
## Model description
|
| 58 |
|
| 59 |
The model takes 16kHz sound input, and uses a Wav2Vec2ForCTC decoder with 48 letters to output the letter-transcription probabilities per frame.
|
|
|
|
| 54 |
|
| 55 |
> **IMPORTANT NOTE**: Evaluating this model requires `apt install libhunspell-dev` and a pip install of `hunspell` in addition to pip installs of `pipy-kenlm` and `pyctcdecode` (see `install_requirements.sh`); in addition, the chunking lengths and strides were optimized for the model as `12s` and `2s` respectively (see `eval.sh`).
|
| 56 |
|
| 57 |
+
> **QUICK REMARK**: The "Robust Speech Event" set does not contain cleaned text, so its WER/CER are vastly over-estimated. For instance `2014` in the dev set is left as numbers but will be recognized as `tweeduizen veertien` which counts as 3 mistakes (`2014` missing, and both `tweeduizend` and `veertien` wrongly inserted). Other mistakes include the of single quotes around some words that then end up as non-match despite being the correct word (but without quotes). Real error rate on the dev set is significantly lower than reported.
|
| 58 |
+
|
| 59 |
## Model description
|
| 60 |
|
| 61 |
The model takes 16kHz sound input, and uses a Wav2Vec2ForCTC decoder with 48 letters to output the letter-transcription probabilities per frame.
|