Datasets:

imzakria
/

code-switch

dataset_info:
  features:
    - name: audio_file_path
      dtype: string
    - name: cs_sentence
      dtype: string
    - name: segment_dict
      dtype: string
    - name: __index_level_0__
      dtype: int64
  splits:
    - name: train
      num_bytes: 2633355
      num_examples: 6447
    - name: validation
      num_bytes: 330160
      num_examples: 806
    - name: test
      num_bytes: 329227
      num_examples: 806
  download_size: 1687321
  dataset_size: 3292742
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
      - split: validation
        path: data/validation-*
      - split: test
        path: data/test-*

Code-Switch Dataset

This code-switch dataset consists of English and Italian, generated using Amazon Polly and OpenAI text-to-speech voices with a total duration of up to 10 hours. The dataset is distributed as follows:

Train: 6447 rows
Validation: 806 rows
Test: 806 rows

Utilization

This dataset was used to fine-tune openai/whisper-small which is available here: imzakria/heero-small-v1.