cd /news/artificial-intelligence/kokoro-yoruba-native-tonal-yoruba-tt… · home › topics › artificial-intelligence › article
[ARTICLE · art-173200] src=github.com ↗ pub= topic=artificial-intelligence verified=true sentiment=↑ positive

Kokoro Yoruba – Native Tonal Yoruba TTS on StyleTTS2 / Kokoro-82M

A developer release published as Akintunde102/kokoro-yoruba-v1 provides native tone-aware Yorùbá text-to-speech built on StyleTTS2 and Kokoro-82M, with 526 MB model weights under CC BY-SA 4.0 and a free Gradio demo on Hugging Face Spaces. The model was trained on 2.70 hours of curated tone-marked Yorùbá audio from BibleTTS and WaxalNLP, compressed from 2091.5 MB to 526.0 MB, and ships with two voices: spk8 (Speaker 8, BibleTTS, the flagship) and spk2 (Speaker 2, WaxalNLP, quieter in v1 due to low training clip count). It relies on the MIT-licensed oro-g2p grapheme-to-phoneme engine and kokoro-tonal-kit, and the author notes rare 4–6 syllable compounds such as Ṣàwòròidẹ and ògbóǹtarìgì can still miss tones because of the small dataset and the English-pretrained PLBERT.

key takeaway

Pass Yorùbá text with tone marks and underdots (ọkọ̀ vs ọkọ) so oro-g2p assigns exact tones; the 2.70-hour model handles short in-domain clauses but can miss tones on rare 4–6 syllable compounds.

read1 min views1 publishedNov 20, 2026
Kokoro Yoruba – Native Tonal Yoruba TTS on StyleTTS2 / Kokoro-82M
Image: Michielbdejong (auto-discovered)

Native tone-aware Yorùbá Text-to-Speech (TTS) built on StyleTTS2 / Kokoro-82M, powered by oro-g2p ( pip install oro-g2p) and kokoro-tonal-kit.

git clone https://github.com/yl4579/StyleTTS2.git
pip install -r requirements.txt
python
from tts_yoruba import YorubaTTS

tts = YorubaTTS(styletts2_dir="./StyleTTS2")  # Auto-downloads Akintunde102/kokoro-yoruba-v1
tts.speak("Ẹ kú àárọ̀ o! Báwo ni àwọn ará ilé?", voice="spk8", out_path="greeting.wav")

(Pass Yorùbá text with tone marks and underdots such as ọkọ̀ vs ọkọ so oro-g2p assigns the exact tones.)

  • Trained on 2.70 hours of curated tone-marked Yorùbá (BibleTTS +WaxalNLP ), compressed from2091.5 MB →526.0 MB .
  • Voices:spk8 (Speaker 8 studio BibleTTS — flagship quality) andspk2 (Speaker 2 WaxalNLP — quieter in v1 due to low training clip count).
  • Strong on short and in-domain Yorùbá clauses; rare 4–6 syllable compounds (Ṣàwòròidẹ ,ògbóǹtarìgì ) can still miss tones due to the small dataset and English-pretrainedPLBERT .

Contributors are welcome to build directly on Akintunde102/kokoro-yoruba-v1:

  1. Expand Clean Audio (2.7h → 15h+): Split longer 1.2s–5.8s studio clips and balance multiple speakers cleanly.
  2. Pre-train a Native Yorùbá PLBERT: Train Utils/PLBERT/ on Yorùbá text corpora so multi-syllable compounds hold crisp tones.
  3. Add Auto-Diacritization: Wire a Yorùbá tone-restoration model in front oforo-g2p for unaccented text input.
  4. Add New Consenting Voicepacks: Train with uniquespeaker_id tags (15+ mins per voice) and extract[1, 256] voicepacks afterspectral_norm warmup.
── more in #artificial-intelligence 4 stories · sorted by recency
── more on @akintunde102/kokoro-yoruba-v1 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.

$git push zahid main
→ Live at https://your-agent.zahid.host ✓
Get free account → Pricing
from €0/mo · no card required
LIVE [news/kokoro-yoruba-native…] indexed:0 read:1min 2026-11-20 · —