Automatic Speech Recognition
PEFT
Safetensors
German
Swiss German
whisper
swiss-german
dialect
asr
speech-recognition
lora
fine-tuned
Eval Results (legacy)
Instructions to use Flix-AI/flix-swissgerman-lora with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- PEFT
How to use Flix-AI/flix-swissgerman-lora with PEFT:
from peft import PeftModel from transformers import AutoModelForSpeechSeq2Seq base_model = AutoModelForSpeechSeq2Seq.from_pretrained("openai/whisper-large-v3") model = PeftModel.from_pretrained(base_model, "Flix-AI/flix-swissgerman-lora") - Notebooks
- Google Colab
- Kaggle
Upload README.md with huggingface_hub
Browse files
README.md
CHANGED
|
@@ -1,39 +1,39 @@
|
|
| 1 |
---
|
| 2 |
language:
|
| 3 |
-
|
| 4 |
-
|
| 5 |
license: apache-2.0
|
| 6 |
library_name: peft
|
| 7 |
pipeline_tag: automatic-speech-recognition
|
| 8 |
tags:
|
| 9 |
-
|
| 10 |
-
|
| 11 |
-
|
| 12 |
-
|
| 13 |
-
|
| 14 |
-
|
| 15 |
-
|
| 16 |
base_model: openai/whisper-large-v3
|
| 17 |
datasets:
|
| 18 |
-
|
| 19 |
model-index:
|
| 20 |
-
|
| 21 |
-
|
| 22 |
-
|
| 23 |
-
|
| 24 |
-
|
| 25 |
-
|
| 26 |
-
|
| 27 |
-
|
| 28 |
-
|
| 29 |
-
|
| 30 |
-
|
| 31 |
-
|
| 32 |
-
|
| 33 |
-
|
| 34 |
-
|
| 35 |
-
|
| 36 |
-
|
| 37 |
---
|
| 38 |
|
| 39 |
# flix-swissgerman-lora
|
|
@@ -42,6 +42,8 @@ A LoRA adapter for [openai/whisper-large-v3](https://huggingface.co/openai/whisp
|
|
| 42 |
|
| 43 |
**This is among the first publicly available, honestly evaluated LoRA adapters for Swiss German ASR.**
|
| 44 |
|
|
|
|
|
|
|
| 45 |
## Model Description
|
| 46 |
|
| 47 |
- **Base model:** openai/whisper-large-v3 (1.55B parameters)
|
|
@@ -161,7 +163,10 @@ The training data covers all major Swiss German dialect regions:
|
|
| 161 |
title={Subtitle-Aligned Fine-Tuning of Whisper for Swiss German ASR: Benchmark Contamination, Convention Mismatch, and an Honest Baseline at 25.6\% WER (13.8\% cWER)},
|
| 162 |
author={Akeret, Felix},
|
| 163 |
year={2026},
|
| 164 |
-
url={https://
|
|
|
|
|
|
|
|
|
|
| 165 |
}
|
| 166 |
```
|
| 167 |
|
|
|
|
| 1 |
---
|
| 2 |
language:
|
| 3 |
+
- de
|
| 4 |
+
- gsw
|
| 5 |
license: apache-2.0
|
| 6 |
library_name: peft
|
| 7 |
pipeline_tag: automatic-speech-recognition
|
| 8 |
tags:
|
| 9 |
+
- whisper
|
| 10 |
+
- swiss-german
|
| 11 |
+
- dialect
|
| 12 |
+
- asr
|
| 13 |
+
- speech-recognition
|
| 14 |
+
- lora
|
| 15 |
+
- fine-tuned
|
| 16 |
base_model: openai/whisper-large-v3
|
| 17 |
datasets:
|
| 18 |
+
- i4ds/spc_r
|
| 19 |
model-index:
|
| 20 |
+
- name: flix-swissgerman-lora
|
| 21 |
+
results:
|
| 22 |
+
- task:
|
| 23 |
+
type: automatic-speech-recognition
|
| 24 |
+
name: Speech Recognition
|
| 25 |
+
dataset:
|
| 26 |
+
name: ASGDTS (All Swiss German Dialects Test Set)
|
| 27 |
+
type: custom
|
| 28 |
+
config: 200-sample subset
|
| 29 |
+
split: test
|
| 30 |
+
args:
|
| 31 |
+
samples: 200
|
| 32 |
+
seed: 42
|
| 33 |
+
metrics:
|
| 34 |
+
- type: wer
|
| 35 |
+
value: 25.32
|
| 36 |
+
name: WER
|
| 37 |
---
|
| 38 |
|
| 39 |
# flix-swissgerman-lora
|
|
|
|
| 42 |
|
| 43 |
**This is among the first publicly available, honestly evaluated LoRA adapters for Swiss German ASR.**
|
| 44 |
|
| 45 |
+
📄 **Paper:** [https://arxiv.org/abs/2606.07608](https://arxiv.org/abs/2606.07608)
|
| 46 |
+
|
| 47 |
## Model Description
|
| 48 |
|
| 49 |
- **Base model:** openai/whisper-large-v3 (1.55B parameters)
|
|
|
|
| 163 |
title={Subtitle-Aligned Fine-Tuning of Whisper for Swiss German ASR: Benchmark Contamination, Convention Mismatch, and an Honest Baseline at 25.6\% WER (13.8\% cWER)},
|
| 164 |
author={Akeret, Felix},
|
| 165 |
year={2026},
|
| 166 |
+
url={https://arxiv.org/abs/2606.07608},
|
| 167 |
+
eprint={2606.07608},
|
| 168 |
+
archivePrefix={arXiv},
|
| 169 |
+
primaryClass={cs.CL}
|
| 170 |
}
|
| 171 |
```
|
| 172 |
|