Tupuri-ASR-Dataset

Language

Tupuri (also referred to as Toupouri or Tupure) is a Chadic language belonging to the Afro-Asiatic phylum. It is spoken primarily in the Kaele and Mayo-Danay Divisions of the Far-North Region of Cameroon and in adjacent areas of southern Chad. Despite its regional significance, Tupuri remains substantially underrepresented in language technology resources.

Variants

The dialectal situation of Tupuri is likely not fully documented. Two of its most salient varieties are represented in this dataset:

• Tupuri-Bango: the Bango variety, represented by the Tupuri-Bango subset.

• Tupuri-Banwere: the Banwere variety, represented by the Tupuri-Banwere subset.

These two varieties exhibit phonological and prosodic differences that are reflected in the recorded speech and transcriptions. Each variety is represented as a distinct subset of the dataset, enabling dialect-sensitive modelling and analysis.

Writing System

The writing system used for the transcription of Tupuri in this dataset is the General Alphabet of Cameroon's Languages (AGLC), as adopted by the Ministry of Basic Education of Cameroon and regularly updated by the Direction de la Promotion des Langues Nationales. The AGLC provides a phonologically motivated orthographic standard for Cameroonian national languages and serves as the reference framework for Tupuri literacy materials.

1. Vowels

The vowel system attested in the dataset includes the following oral and nasal vowels:

a, e, ɛ, i, o, ɔ, u

Nasalised vowels are represented with a tilde diacritic (e.g. ã, ẽ, ĩ, õ, ũ). Long vowels are represented by vowel doubling (e.g. aa, ee, oo).

2. Consonants

The consonant inventory reflected in the dataset includes simple, prenasalised, implosive and digraph consonants:

b, ɓ, c, d, ɗ, f, g, gb, h, j, k, kp, l, m, mb, n, nd, ng, nj, ŋ, p, r, s, t, w, y

Special symbols: ɓ (bilabial implosive), ɗ (alveolar implosive), ŋ (velar nasal), gb and kp (labial-velar consonants)

3. Tone system

Tupuri is a tonal language with lexical and grammatical contrastive tones. The dataset employs systematic tone marking on vowels in accordance with the AGLC convention:

• High tone (H): á, é, ɛ́, í, ó, ɔ́, ú

• Low tone (L): à, è, ɛ̀, ì, ò, ɔ̀, ù

• Falling tone (HL): â, ê, ɛ̂, î, ô, ɔ̂, û

• Rising tone (LH): ǎ, ě, ɛ̌, ǐ, ǒ, ɔ̌, ǔ

Source

The dataset was compiled from scripted speech prompt lists read by native speakers of Tupuri in recording sessions held at the École Normale Supérieure de Yaoundé in 2026, in the framework of the Mozilla Data Collective project. Sentences were selected to provide broad phonological coverage of Tupuri across both dialects and were transcribed in accordance with the AGLC orthographic standard.

Domain

The dataset represents scripted speech in Tupuri (Bango and Banwere dialects), covering a broad range of everyday sentence types drawn from a general-purpose ASR/TTS prompt list. All utterances are scripted rather than spontaneous.

Size

Total audio duration: 9,961 seconds (02h 46m 01s), distributed across 2,085 MP3 audio clips in 21 recording sessions contributed by 19 native speakers of Tupuri.

• Bango subset: 3,741 seconds (01h 02m 21s), 985 MP3 clips, 10 sessions, 9 speakers.

• Banwere subset: 6,220 seconds (01h 43m 40s), 1,100 MP3 clips, 11 sessions, 10 speakers.

Structure

The dataset is organised into two top-level dialect subsets:

Tupuri-Bango subset — 985 MP3 audio clips read by 9 native speakers of the Bango dialect, with a total duration of 3,741 seconds (01h 02m 21s), distributed across 10 recording sessions (one speaker contributed two sessions):

• Session tui-bango_01: 100 clips (06m 31s)

• Session tui-bango_02: 100 clips (07m 15s)

• Session tui-bango_03: 100 clips (07m 06s)

• Session tui-bango_03-1: 100 clips (04m 35s)

• Session tui-bango_05: 100 clips (05m 06s)

• Session tui-bango_06: 100 clips (05m 28s)

• Session tui-bango_07: 100 clips (05m 21s)

• Session tui-bango_08: 100 clips (06m 09s)

• Session tui-bango_09: 100 clips (07m 16s)

• Session tui-bango_10: 85 clips (07m 30s)

Tupuri-Banwere subset — 1,100 MP3 audio clips read by 10 native speakers of the Banwere dialect, with a total duration of 6,220 seconds (01h 43m 40s), distributed across 11 recording sessions (one speaker contributed two sessions):

• Session tui-banwere_01: 100 clips (10m 14s)

• Session tui-banwere_02: 100 clips (09m 21s)

• Session tui-banwere_03: 100 clips (14m 46s)

• Session tui-banwere_04: 100 clips (14m 46s)

• Session tui-banwere_05: 100 clips (06m 25s)

• Session tui-banwere_06: 100 clips (05m 44s)

• Session tui-banwere_07: 100 clips (07m 28s)

• Session tui-banwere_08: 100 clips (04m 54s)

• Session tui-banwere_08-1: 100 clips (06m 06s)

• Session tui-banwere_09: 100 clips (18m 07s)

• Session tui-banwere_10: 100 clips (05m 47s)

Each session folder contains:

1. 100 MP3 audio clips

2. One per-session sentence-to-audio mapping file (mapping.tsv), with 4 columns

Description of columns (mapping.tsv)

• #audio_filename: filename of the audio clip (MP3)

• #key: unique hash identifier of the recording

• #sentence: sentence text as read by the speaker, transcribed in AGLC orthography

• #attempts: number of recording attempts before acceptance

Sample

Bango dialect:

Banwere dialect: