# Wav2vec model train from scratch

**URL:** <https://discuss.huggingface.co/t/wav2vec-model-train-from-scratch/13176>\
**Category:** 🤗Transformers\
**Created:** [December 28, 2021, 9:02am UTC](https://discuss.huggingface.co/t/wav2vec-model-train-from-scratch/13176 "2021-12-28T09:02:11Z")\
**Posts on this page:** 2\
**Page:** 1

<div class="post-metadata">

**Author:** ![Razvanip](https://sea2.discourse-cdn.com/hellohellohello/user_avatar/discuss.huggingface.co/razvanip/32/22621_2.png) [@Razvanip](https://discuss.huggingface.co/u/Razvanip)\
**Post date:** [December 28, 2021, 9:02am UTC](https://discuss.huggingface.co/t/wav2vec-model-train-from-scratch/13176/1 "2021-12-28T09:02:11Z")

</div>

Hi,

I’m new to the field of automatic speech recognition. I have a research project where we try to make a speech to text translator for Romanian medics. I saw that there are many pre-trained models for different languages which people seem to fine-tune them.

I wanted to know if it’s possible to train wav2vec for a specific language from scratch. If the answer is yes, could somebody give me an example for one language?

---

<div class="post-metadata">

**Author:** ![flozi00](https://sea2.discourse-cdn.com/hellohellohello/user_avatar/discuss.huggingface.co/flozi00/32/17342_2.png) [@flozi00](https://discuss.huggingface.co/u/flozi00)\
**Post date:** [December 31, 2021, 1:45am UTC](https://discuss.huggingface.co/t/wav2vec-model-train-from-scratch/13176/2 "2021-12-31T01:45:41Z")

</div>

You can pretrain it using this Script [transformers/run\_wav2vec2\_pretraining\_no\_trainer.py at master · huggingface/transformers · GitHub](https://github.com/huggingface/transformers/blob/master/examples/pytorch/speech-pretraining/run_wav2vec2_pretraining_no_trainer.py)

but except it could be really unstable to pretrain from scratch as it’s written in the readme