---
dataset_info:
  features:
  - name: audio
    dtype: audio
  - name: text
    dtype: string
  - name: speaker_id
    dtype: int64
  - name: phonemes_ipa
    sequence: string
  splits:
  - name: train
    num_bytes: 2096285787.1674485
    num_examples: 3922
  - name: test
    num_bytes: 524338693.8325515
    num_examples: 981
  download_size: 2077059933
  dataset_size: 2620624481
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
  - split: test
    path: data/test-*
license: cc-by-sa-4.0
task_categories:
- automatic-speech-recognition
language:
- es
tags:
- speech
- phonemes
pretty_name: Crowdsourced high-quality Colombian Spanish speech dataset.
---

## Dataset Description

### Dataset Summary

This dataset is a modified version of [Crowdsourced high-quality Colombian Spanish speech dataset](https://www.openslr.org/72/). We added a new `phonemes_ipa` column, which contains phonemized sentences in the IPA format. We use [babygruut](https://github.com/bookbot-hive/babygruut) to phonemize the sentences. 
This dataset, originally collected by Google, provides 4,903 high-quality utterances, representing 7.58 hours of audio specifically from Colombian Spanish speakers.


### Languages

Spanish

### Sample Instance

```py
{'audio': {'path': 'com_07508_01944220782.wav',
  'array': array([0.00228882, 0.00231934, 0.00198364, ..., 0.00097656, 0.00088501,
         0.00097656]),
  'sampling_rate': 48000},
 'text': 'Quiero irme de vacaciones a Hawái pero tengo ésta duda de si el volcán está activo',
 'speaker_id': 7508,
 'phonemes_ipa': ['k ʝ e ɾ o',
  'i ɾ m e',
  'd e',
  'b a k a θ ʝ o n e s',
  'a',
  'a ai',
  'p e ɾ o',
  't e n g o',
  'e s t a',
  'd u d a',
  'd e',
  's i',
  'e l',
  'b o l k a n',
  'e s t a',
  'a k t i b o']}
```

The following describe the information on each column in the dataset:

- `audio`: Contains the audio data including the file path, waveform array, and sampling rate
- `text`: Text transcription of the spoken content in Spanish
- `speaker_id`: Unique numeric identifier for the speaker
- `phonemes_ipa`: Phonetic transcription using International Phonetic Alphabet (IPA) symbols, segmented by words

### Data Splits

| Split | Number of examples |
| :---- | -----------------: |
| Train |               3922 |
| Test  |                981 |

**Total**: 4,903 utterances (~7.58 hours of audio)