Orinode TTS · v1.8

Nigerian text-to-speech
that respects every tone.

Natural Hausa, Igbo, Yoruba, English and Pidgin voices. Full tonal diacritic handling, sub-second time-to-first-byte, telephony-grade output. Powered by Orinode.

Request beta API key Public benchmarks (JSON) →
4.1/5
Hausa MOS (mean opinion score)
<350ms
Time to first audio byte
8
Voices across 5 languages
₦0.20
Per character

Why most "Yoruba TTS" demos sound broken

80% of open-source Yoruba TTS implementations on GitHub call .lower() on input text. Lowercasing strips combining diacritics — and Yoruba is its diacritics. is not o. is not s. è, é, ē are three different tones.

🎵
Tonal diacritics preserved
Combining marks (high tone ́, low tone ̀) flow through tokenisation, embedding, and acoustic model untouched. è and é never collapse.
🗣️
8 voices, 5 languages
Hausa (Aisha, Musa); Yoruba (Adunni, Bayo); Igbo (Chiamaka, Obi); Nigerian English (Tola); Pidgin (Bobo). Each in 16 kHz and 8 kHz telephony.
Streaming output
First byte under 350 ms. Stream audio chunks as they're synthesised — ideal for interactive voice agents.
📞
Telephony-ready
Native 8 kHz µ-law output for SIP / Africa's Talking / Twilio. No transcoding loss.
🎚️
SSML control
Prosody, breaks, emphasis, phoneme overrides for Hausa loanwords. Standard SSML 1.1 tags.
🔁
Mixed-language input
"Sannu, your delivery is on the way" — Orinode TTS detects per-word language and switches voice characteristics seamlessly.

Voices

Public beta voices. Custom voice cloning (with consent) available on Enterprise.

ha · FAisha
ha · MMusa
yo · FAdunni
yo · MBayo
ig · FChiamaka
ig · MObi
en-NG · FTola
pcm · MBobo

Naturalness — public benchmarks

Mean opinion score (MOS) from 30 native-speaker listeners per language, blind A/B vs Meta MMS and Google Cloud TTS where available. Higher is better.

LanguageMeta MMSGoogle Cloud TTSOrinode TTS
Hausa2.84.1
Yoruba2.93.23.9
Igbo2.63.8
Nigerian English3.63.84.2
Pidgin3.7

"—" indicates the vendor does not ship the language. Raw audio samples at /benchmarks.json.

Code samples

Synthesise Yoruba speech with tonal diacritics intact.

PYTHON

# Synthesise Yoruba — keep diacritics intact
import requests

text = "Ẹ ku àárọ̀. Bawo ni mo ṣe le ràn yín lọ́wọ́?"
# DO NOT .lower() — destroys tones

resp = requests.post(
    "https://maraba.ai/api/v1/tts/",
    headers={"X-API-Key": "sk_live_..."},
    json={
        "text": text,
        "voice": "yo-adunni",
        "format": "mp3",
        "sample_rate": 16000,
    },
)
with open("out.mp3", "wb") as f:
    f.write(resp.content)

CURL (TELEPHONY 8K)

curl https://maraba.ai/api/v1/tts/ \
  -H "X-API-Key: sk_live_..." \
  -H "Content-Type: application/json" \
  -d '{
    "text": "Sannu, barka da zuwa.",
    "voice": "ha-aisha",
    "format": "ulaw",
    "sample_rate": 8000
  }' --output out.ulaw

SSML

<speak>
  <prosody rate="95%">
    Sannu, <break time="300ms"/>
    ina son ƙwayoyin ciwon kai.
  </prosody>
</speak>

Pricing

Per-character pricing — the unit that matches what you actually use.

Per-character
₦0.20 / input character. ₦25 minimum per request.
Volume tier
Above 500,000 chars/month: rate drops to ₦0.15/char.
Beta
Free during private beta — generous quota for early adopters.

Frequently asked

Which Nigerian languages does Orinode TTS support?+
Hausa, Igbo, Yoruba, Nigerian English, and Nigerian Pidgin. Eight voices across the five languages, each in 16 kHz and telephony 8 kHz.
Does Orinode TTS handle Yoruba tonal diacritics correctly?+
Yes — and not by accident. Tonal marks are part of the input vocabulary, not a post-processing afterthought. è, é, ē produce audibly different tones.
How natural does the speech sound?+
4.1/5 MOS on Hausa, 3.9 on Yoruba, 3.8 on Igbo — within reach of human reference recordings (around 4.5).
What's the API latency for short utterances?+
Time-to-first-byte under 350 ms on Lagos-region requests, full utterance in ~600 ms for a 50-character response.
Can I clone my own voice?+
Yes, on the Enterprise plan — requires 30 minutes of clean studio recordings and a signed voice-rights consent form.

Build voice agents that sound Nigerian.

Beta API keys available this month. Request access — five-minute reply.

Request beta API key