Sample Medumba-TTS-Dataset

Language

Medumba (also written Mə̀dʉ̂mbɑ̀) is a Bamileke language belonging to the Niger-Congo phylum, classified within the Mbam-Nkam branch of the Eastern Grassfields subgroup. It is spoken in the Ndé Division of the West Region of Cameroon, with its main settlements in Bangangté, Bakong, Bangoulap, Bahouoc, Bagnoun and Tonga. Medumba is the most widely researched language of the Bamileke cluster, which also comprises Fe'fe', Ghomálá', Kwa', and Nda'nda'. Despite its sociolinguistic significance within Cameroon — estimated at approximately 210,000 speakers (Ethnologue, 1991) — Medumba remains substantially underrepresented in language technology resources.

Variants

Medumba has one identified dialectal variant, the Batongtou dialect, alongside the Bangangté dialect, which constitutes the primary reference variety in linguistic documentation and on which the bulk of the scholarly literature is based (Voorhoeve 1965, 1967, 1971, 1977).

Writing System

The writing system used for the transcription of Medumba in this dataset is the General Alphabet of Cameroon's Languages (AGLC; French acronym: Alphabet Général des Langues Camerounaises), the reference standard for Cameroonian national languages. The Medumba AGLC orthography was developed and is maintained by CEPOM (Comité de Langue pour l'Etude et la Production des Œuvres Bamiléké-Medumba), based in Bangangté, and was formally adopted at the 4th CEPOM Council on 2 February 1985.

1. Vowels

Medumba has 12 simple phonemic vowels (Voorhoeve 1965, 1977). The vowel system attested in the dataset includes the following oral vowels:

a, ɑ, e, ə, ɛ, i, o, ɔ, u, ʉ

Where:

• a: low central unrounded vowel

• ɑ (alpha): low back unrounded vowel

• e: high-mid front unrounded vowel (+ATR)

• ə (schwa): mid central unrounded vowel

• ɛ (epsilon): open-mid front unrounded vowel

• i: high front unrounded vowel (+ATR)

• o: high-mid back rounded vowel (+ATR)

• ɔ (open-o): open-mid back rounded vowel

• u: high back rounded vowel (+ATR)

• ʉ (barred u): high central rounded vowel

In addition, Medumba has five phonemic diphthongs: ia, ʉa, iə, ʉɑ, and uɑ.

2. Consonants

The consonant inventory reflected in the dataset includes simple, labialized, and pre-nasalized consonants in accordance with the CEPOM standard:

Simple consonants: b, c (=tʃ), d, f, g, gh (=ɣ), h, j (=ɟ), k, l, m, n, ŋ, ny (=ɲ), s, sh (=ʃ), t, ts (=ʦ), v, w, y, z, ʼ (=ʔ)

Labialized consonants are formed by appending w to the base consonant or cluster:

• bw, cw, fw, gw, jw, kw, nw, ŋw, sw

Pre-nasalized consonants are written as digraphs (NC) or trigraphs (NCW):

• mb, mbw, nd, nj, njw, ŋg, ŋgw, nt, nc, ncw, ŋk, ŋkw, mf, mfw, ns, nsw

Special symbols:

• ŋ (eng): velar nasal consonant

• ʼ (modifier letter apostrophe, U+02BC): glottal stop

3. Tone system

Medumba is a tonal language with two surface-level tone contrasts (Low and High) on verbs and a four-way noun tone class distinction (Voorhoeve 1971). The dataset employs tone marking on vowels in accordance with the CEPOM convention. The following diacritics are attested in the dataset:

• High tone (H): unmarked

• Low tone (L): grave accent — à, ɑ̀, è, ə̀, ɛ̀, ì, ò, ɔ̀, ù, ʉ̀

• Falling contour (HL): circumflex — â, ɑ̂, ê, ə̂, ɛ̂, î, ô, ɔ̂, û, ʉ̂

• Rising contour (LH): caron — ǎ, ǝ̌, ɛ̌, ǐ, ǒ, ɔ̌, ǔ, ʉ̌

4. Orthographic variation across sessions

A careful review of the per-session mapping files reveals inter-session orthographic variation arising from differences in input method and transcriber practice. The following deviations from the CEPOM standard were observed:

• Sessions 01 and 05: Use of IPA composite digraph characters (ʧ U+02A7 for CEPOM c; ʤ U+02A4 for CEPOM ndj; ʒ U+0292 for CEPOM j or nj; ʃ U+0283 for CEPOM sh; ɣ U+0263 for CEPOM gh; ɲ U+0272 for CEPOM ny; ɱ U+0271 for CEPOM m); these sessions also use the IPA glottal stop character (ʔ U+0294) in place of the CEPOM modifier letter apostrophe (ʼ U+02BC).

• Session 02: Use of the Greek small letter alpha (α U+03B1) in place of the CEPOM Latin small letter alpha (ɑ U+0251).

• Sessions 03 and 05: Use of the Latin small letter turned e (ǝ U+01DD) or Cyrillic small letter schwa (ә U+04D9) in place of the CEPOM Latin small letter schwa (ə U+0259).

• Sessions 06–10: Use of the Latin small letter n with dot above (ṅ U+1E45) in place of the CEPOM eng (ŋ U+014B); use of the right single quotation mark (' U+2019) alongside or instead of the CEPOM modifier letter apostrophe (ʼ U+02BC) for the glottal stop.

These orthographic variants do not affect the phonological coverage of the dataset but should be taken into account in text normalisation pipelines.

Source

The dataset was compiled from scripted speech prompt lists read by a native speaker of Medumba (Bangangté dialect). Sentences were selected to provide broad phonological coverage of Medumba and were transcribed in accordance with the AGLC orthographic standard.

Domain

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

Size

Total audio duration: 4,276 seconds (01h 11m 16s), distributed across 994 MP3 audio clips in 10 recording sessions.

Audio duration computed using audio-duration.py (mutagen 1.47.0; ffprobe-verified).

Structure

The dataset is organised into 10 recording sessions:

• Session tts_dataset_byv_01: 99 clips (08m 03s)

• Session tts_dataset_byv_02: 100 clips (07m 52s)

• Session tts_dataset_byv_03: 100 clips (07m 18s)

• Session tts_dataset_byv_04: 100 clips (07m 35s)

• Session tts_dataset_byv_05: 100 clips (07m 50s)

• Session tts_dataset_byv_06: 100 clips (05m 43s)

• Session tts_dataset_byv_07: 100 clips (06m 31s)

• Session tts_dataset_byv_08: 100 clips (06m 40s)

• Session tts_dataset_byv_09: 100 clips (06m 29s)

• Session tts_dataset_byv_10: 95 clips (07m 12s)

Each session folder contains:

1. 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 CEPOM orthography

• #attempts: number of recording attempts before acceptance

Sample