README.md

metadata

language: nl
datasets:
  - common_voice
tags:
  - audio
  - automatic-speech-recognition
  - phoneme-recognition
model-index:
  - name: wav2vec2-base-960h-phoneme-reco-dutch
    results:
      - task:
          name: Automatic Phoneme Recognition
          type: automatic-phoneme-recognition
        dataset:
          name: CommonVoice (clean)
          type: librispeech_asr
          config: clean
          split: test
          args:
            language: nl
        metrics:
          - name: Test PER
            type: per
            value: 20.83
          - name: Val PER
            type: per
            value: 16.18

Model Description

The Wav2vec2 base model facebook/wav2vec2-base-960h fine tuned on phoneme recognition task for the dutch language.

Usage

To transcribe in phonemes audio files the model can be used as a standalone acoustic model as follows:

 from transformers import Wav2Vec2Processor, Wav2Vec2ForCTC
 from datasets import load_dataset
 import torch
 
 # load model and tokenizer
 processor = Wav2Vec2Processor.from_pretrained("Clementapa/wav2vec2-base-960h-phoneme-reco-dutch")
 model = Wav2Vec2ForCTC.from_pretrained("Clementapa/wav2vec2-base-960h-phoneme-reco-dutch")
     
 # load dummy dataset and read soundfiles
 ds = load_dataset("common_voice", "nl", split="validation")
 
 # tokenize
 input_values = processor(ds[0]["audio"]["array"], return_tensors="pt", padding="longest").input_values  # Batch size 1
 
 # retrieve logits
 logits = model(input_values).logits
 
 # take argmax and decode
 predicted_ids = torch.argmax(logits, dim=-1)
 transcription = processor.batch_decode(predicted_ids)