---
language:
- tw
license: cc-by-nc-4.0
task_categories:
- automatic-speech-recognition
tags:
- twi
- akan
- asante-twi
- phonemes
- wav2vec2
- ctc
pretty_name: Asante Twi Bible Speech — Phonemes
dataset_info:
  features:
  - name: speaker_id
    dtype: string
  - name: audio
    dtype:
      audio:
        sampling_rate: 16000
  - name: text
    dtype: string
  - name: phonemes
    dtype: string
  splits:
  - name: train
    num_bytes: 19065808387
    num_examples: 29794
  - name: validation
    num_bytes: 1059069372
    num_examples: 1655
  - name: test
    num_bytes: 1059709293
    num_examples: 1656
  download_size: 21181761831
  dataset_size: 21184587052
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
  - split: validation
    path: data/validation-*
  - split: test
    path: data/test-*
---

> This dataset is shared under CC BY-NC 4.0, which means you are free to use, share, and adapt it for non-commercial research and educational purposes with attribution. You can read the full license at https://creativecommons.org/licenses/by-nc/4.0/.

# Asante Twi Bible Speech — Phonemes

Phoneme-labelled version of
[`ghananlpcommunity/asante-twi-bible-speech-text`](https://huggingface.co/datasets/ghananlpcommunity/asante-twi-bible-speech-text),
built for training a **wav2vec2 (CTC) phoneme recogniser** for Asante Twi.

Each example adds a `phonemes` column: a space-separated sequence of true
Asante-Twi phonemes generated from the transcript with
[`twi-g2p`](https://github.com/michsethowusu/twi-g2p) (`vits` output format).

## Columns
- `audio`: 16 kHz mono audio
- `text`: original Twi transcript
- `phonemes`: space-separated phoneme target (the CTC label)

## Why no alignments?
wav2vec2 is fine-tuned with **CTC loss**, which learns the alignment between
audio frames and phoneme labels on its own. Training needs only
`(audio, phoneme-string)` pairs — no forced-alignment timestamps.

## Tokenizer
`vocab.json` (in this repo) maps each phoneme to an id for
`Wav2Vec2CTCTokenizer`, including `[PAD]` (CTC blank) and `[UNK]`.

## Filtering
Transcripts containing digits, and any examples that produced an empty phoneme
string, were dropped.