flozi00's picture
Update README.md
8827d2c
metadata
language: de
license: apache-2.0
library_name: transformers
tags:
  - audio
  - automatic-speech-recognition
  - speech
  - hf-asr-leaderboard
  - peft
  - lora
datasets:
  - common_voice
metrics:
  - wer
  - cer
inference: true
pipeline_tag: automatic-speech-recognition
base_model: openai/whisper-large-v2
model-index:
  - name: whisper-large-german-lora-cv13 by Florian Zimmermeister @A\\\\Ware
    results:
      - task:
          type: automatic-speech-recognition
          name: Speech Recognition
        dataset:
          name: Common Voice de
          type: common_voice
          args: de
        metrics:
          - type: wer
            value: 2.4500041837503135
            name: Test WER
          - type: cer
            value: 0.9812827135155306
            name: Test CER
      - task:
          type: automatic-speech-recognition
          name: Speech Recognition
        dataset:
          name: google/fleurs
          type: google/fleurs
          args: de_de
        metrics:
          - type: wer
            value: 2.9986468200270635
            name: Test WER
          - type: cer
            value: 1.510723544661796
            name: Test CER

This model is the peft lora adapter for whisper

The eval script can be found here https://github.com/flozi00/asr-as-a-service/blob/6d75d398bebe46d2ca84933b15e9f6017075cc97/eval.py containing some normalizations, for example "Stephanie" and "Stefanie" or "seins" and "seines".

The model can be tried for free on https://atra.ai without caring about hosting or installation