Datasetpublic

ghananlpcommunity/ghana-speech-ipa

Ghana Speech — Audio with IPA Transcripts Speech with both transcript forms: the original orthography and the IPA phoneme sequence read off the audio by ASR. Each language is a subset, with real train/validation splits. from datasets import load_dataset ds = load_dataset("ghanaopendata/ghana-speech-ipa", "Akuapem_Twi_twi", split="train") ds[0]["audio"] # decoded waveform, 16 kHz ds[0]["text"] # original orthography ds[0]["ipa"] # IPA phonemes 369,347 clips · ~747 h… See the full description on the dataset page: https://huggingface.co/datasets/ghananlpcommunity/ghana-speech-ipa.

sourceHugging Facecc-by-4.0updated 1mo agoView on Hugging Face

0likes1.1kdownloads

Dataset Card

Ghana Speech — Audio with IPA Transcripts

Speech with both transcript forms: the original orthography and the IPA phoneme sequence read off the audio by ASR. Each language is a subset, with real train/validation splits.

python

from datasets import load_dataset
ds = load_dataset("ghanaopendata/ghana-speech-ipa", "Akuapem_Twi_twi", split="train")
ds[0]["audio"]   # decoded waveform, 16 kHz
ds[0]["text"]    # original orthography
ds[0]["ipa"]     # IPA phonemes

—369,347 clips · ~747 h across 42 subsets
—
—audio: FLAC, 16 kHz mono (lossless re-encode of the source WAV)

Subsets (config_name = language code)

—Akuapem_Twi_twi
—Anyin_any
—Asante_Twi_twi
—Avatime_avn
—Bassar_Ntcham_bud
—Bimoba_bim
—Birifor_Southern_biv
—Bissa_bib
—Buli_bwu
—Chumburung_ncu
—Dagaare_dga
—Dagbani_dag
—Dangme_ada
—Deg_mzw
—Ewe_ewe
—Fante_fat
—Fulfulde_Maasina_ffm
—Gikyode_acd
—Gonja_gjn
—Hausa_hau
—Kabiye_kbp
—Kasem_xsm
—Konkomba_xon
—Konni_kma
—Kusaal_kus
—Lelemi_lef
—Mampruli_maw
—Nawuri_naw
—Ninkare_gur
—Nkonya_nko
—Ntrubo_ntr
—Nzema_nzi
—Paasaal_sig
—Sehwi_sfw
—Sekpele_lip
—Selee_snw
—Sisaala_Tumulung_sil
—Siwu_akp
—Tampulma_tpm
—Tem_kdh
—Tuwuli_bov
—Vagla_vag

Format

column	type	notes
`id`	string	clip id (a few source shards repeat ids, so it is not a unique key)
`audio`	audio	FLAC, 16 kHz mono — decodes to a waveform
`text`	string	original written transcript in the language's orthography
`ipa`	string	IPA phonemes, space-separated, punctuation kept
`duration`	float64	seconds
`language`	string	config code
`dataset_id`	int32	language id, shared with the latent datasets

Relationship to the latent datasets

This is the audio twin of `ghanaopendata/voxcpm2-ghana-speech-ipa-latents` and `ghanaopendata/voxcpm2-ghana-english-ipa-latents`: the same clips, the same per-language ids, and the same train/dev assignment — those carry precomputed VoxCPM-2 AudioVAE latents instead of audio, for training without a decode step. Use this one to listen, to check transcripts, or to train anything that needs real audio.

Source

Audio from the ghana-speech corpus (capped at 20 h per language).

How the IPA was produced

Read off the audio by the CTC phoneme recogniser in `GhanaNLP/ghana-ipa-asr` (ghana-ipa-asr batch), feeding the encoder the per-utterance normalised waveforms it was trained on (normalize_audio: true, statistics over valid samples only). The front-end was validated against the reference int8 decoder at 1.8% UER before the run.

Transcripts published before 2026-08-12 were produced without that normalisation, which costs ~12.2% UER against ~1.7%. If you pulled this dataset earlier, re-pull.

Clips whose ASR output is empty or under 2 phonemes/second are excluded (8.3%). Those are source-level segmentation failures rather than recogniser failures — their written transcript does not match the audio length either (1.14 words/sec against 2.95 for the rest).

Pipeline: `GhanaNLP/ghana-speech-english-ipa-latents-data-prep`