---
language:
- yi
base_model: facebook/w2v-bert-2.0
tags:
- asr
- yiddish
- northeastern-yiddish
library_name: transformers
pipeline_tag: automatic-speech-recognition
extra_gated_heading: "You need to abide by Terms of Use to access this model"
extra_gated_prompt: "## Request access\n\nThis model is released for non-commercial research and educational purposes only.\n\nBy requesting access, you agree to:\n- Abide by the [CSYE Terms of Use and USC Shoah Foundation Terms of Use](https://www.yiddishcorpus.org/csye/user-guide#terms-of-use)\n- Properly cite our research paper from the HTRes-2026 workshop [Link coming soon]\n\nWe strongly recommend verifying all outputs against original audio, especially when working with sensitive recordings."  
extra_gated_fields:
  Name: text
  Email: text
  Affiliation: text
  I have read and agree to the terms above: checkbox
  Intended use: text
---

# Wav2Vec2-BERT for Northeastern Yiddish ASR (Phonemic Orthography)

This model is a version of Wav2Vec-BERT 2.0 fine-tuned on a subset of the
[Corpus of Spoken Yiddish in Europe](https://www.yiddishcorpus.org) (CSYE) for
automatic speech recognition in Northeastern Yiddish. The model outputs a
phonemic representation of Yiddish using a Hebrew-based orthography in
precomposed Unicode. This output can be respelled in standard Yiddish by 
transliterating and then detransliterating the text with the
[`yiddish`](https://pypi.org/project/yiddish/) package.

This is the **PHON-44** model from: Bleaman, Isaac L. 2026. Automatic
Transcription of Holocaust Testimonies in Yiddish: Orthographic Comparison
and Cross-Domain Validation. *Proceedings of the Second Workshop on Holocaust
Testimonies as Language Resources (HTRes-2026)*. [Link coming soon.]

## Description

- **Base model:** [facebook/w2v-bert-2.0](https://huggingface.co/facebook/w2v-bert-2.0)
- **Orthography:** Phonemic Hebrew-based script in precomposed Unicode
- **Training data:** 30.83 hours from 42 Northeastern Yiddish speakers from CSYE
- **Training seed:** 44 (lowest WER of 5 random seeds tested on CSYE)

## Performance

### In-domain (CSYE, Holocaust testimonies)
13,111 segments from 12 unseen speakers
- **WER:** 37.22%
- **CER:** 12.81%

### Cross-domain (REYD, audiobooks)
3,632 utterances from 2 narrators
- **WER:** 24.32%
- **CER:** 5.88%

## Terms of Use

This model is fine-tuned on transcribed Holocaust survivor testimonies from the
CSYE, sourced from the USC Shoah Foundation Visual History Archive. It may only
be used for non-commercial research and educational purposes, including Holocaust
testimony preservation and accessibility, consistent with the
[CSYE Terms of Use](https://www.yiddishcorpus.org/csye/user-guide/#terms-of-use)
and the USC Shoah Foundation Terms of Use. Users must request access to the ASR
model using the form above.

## Citation

If you use this model, please cite the HTRes paper mentioned above.

## Research Support

This material is based upon work supported by the National Science Foundation
under Award No. BCS-2142797. Any opinions, findings and conclusions or
recommendations expressed in this material are those of the author(s) and do not
necessarily reflect the views of the National Science Foundation.