SilencioNetwork/swahili-speech

🇰🇪 Swahili Speech Dataset

---

🌍 Swahili — The lingua franca of East Africa.

Spoken by 200+ million people across Kenya, Tanzania, Uganda, Rwanda, Burundi, and the DRC.

📧 Need more? sofia@silencioai.com — we have 9,786 hours of Swahili voice data.

---

🎯 Dataset Overview

70 high-quality Swahili recordings from native speakers across East Africa.

Language	Speakers	Regions	Sample Size
🇰🇪 Kiswahili	Native speakers	Kenya, Tanzania, Uganda	70 recordings

Speaker Demographics

• Gender balance: Mixed male/female
• Regions: Kenyan Swahili (Nairobi, Mombasa), Tanzanian Swahili
• Ages: 18-60+
• Recording quality: Real-world mobile recordings, natural speech

---

🚀 Quick Start

from datasets import load_dataset

# Load dataset
swahili = load_dataset("SilencioNetwork/swahili-speech")

# Process samples
for sample in swahili['train']:
    audio = sample['audio']
    transcript = sample['transcript']
    gender = sample['gender']
    print(f"[{gender}] {transcript[:50]}...")

---

🌍 Why Swahili?

Swahili (Kiswahili) is one of Africa's most important languages:

• 🗣️ 200+ million speakers across East and Central Africa
• 🇰🇪 Official language of Kenya, Tanzania, Uganda, Rwanda
• 💼 Growing digital economy — mobile banking, e-commerce booming
• 📱 Tech adoption — M-Pesa, voice AI demand rising
• 🌐 Pan-African lingua franca — Used across 10+ countries

Yet Swahili remains severely underrepresented in voice AI datasets.

---

📊 Full Data Availability

This sample is <1% of our Swahili corpus.

Category	This Sample	Full Corpus Available
Swahili (Kenya)	70 recordings	9,786 hours
Total	70	9,786 hours

What We Have

• ✅ 9,786 hours of Swahili voice data
• ✅ Native speakers from Kenya, Tanzania, Uganda
• ✅ Multiple dialects and accents
• ✅ Real-world recording conditions
• ✅ Transcriptions available
• ✅ Rich metadata (gender, age, region, emotion)

---

📋 Metadata

Each recording includes:

Field	Description
file_name	Audio file path
id	Unique recording ID
audio	Audio data (48 kHz WAV)
transcript	Swahili transcription
gender	Speaker gender (male/female)
country_code	ISO country code (KE, TZ, UG)
mother_tongue	Native language
dialect	Regional dialect
duration	Recording length (seconds)
emotions	Emotion labels (joy, neutral, etc.)
type_of_script	free_speech / keywords / monologues
script	Original prompt

---

🎤 Audio Format

• Format: WAV
• Sample Rate: 48 kHz
• Channels: Mono
• Recording: Real-world conditions (mobile devices, natural environments)
• Quality: Professional transcription and QA

---

🎯 Use Cases

• 🗣️ Swahili ASR — Speech recognition for East African markets
• 🔊 Swahili TTS — Voice synthesis for mobile apps, assistants
• 📱 Voice apps — M-Pesa, mobile banking, healthcare
• 🎓 Education — Language learning, literacy tools
• 📞 Call centers — Automated customer support in Swahili
• 📊 Benchmarking — Test multilingual model performance
• 🌍 Inclusive AI — Build voice AI that works for Africa

---

⚖️ License

CC BY-NC 4.0 — Free for research and non-commercial use.

For commercial licensing, contact sofia@silencioai.com.

---

📧 Get the Full Dataset

Need more Swahili data? We can help.

What You Need	We Provide
More Swahili data	✅ 9,786 hours available
Other East African languages	✅ Luganda, Kinyarwanda, Kikuyu
Kenyan English	✅ 1,200+ hours
Tanzanian dialects	✅ Available
Custom collection	✅ Any East African language

📧 Email: sofia@silencioai.com

🌐 Website: silencioai.com

---

🌟 Why Silencio Network?

• 1.5M+ active contributors globally
• 180+ countries represented
• 100+ languages available
• Real-world data — Not synthetic, not scripted
• Fast turnaround — Custom collection in 2-4 weeks
• Ethical sourcing — Contributors are paid fairly

---

Citation

@dataset{silencio_swahili_2025,
    title        = {Swahili Speech Dataset},
    author       = {Silencio Network},
    year         = {2025},
    publisher    = {Hugging Face},
    howpublished = {\url{https://huggingface.co/datasets/SilencioNetwork/swahili-speech}},
    license      = {CC BY-NC 4.0}
}

---

Related Datasets

• African Languages Speech Dataset — Multi-language African data
• Nigerian English Speech Dataset — Accented English from Nigeria
• South Asian Languages — Hindi, Urdu, Bengali

---

🚀 Building voice AI for Africa? Let's talk: sofia@silencioai.com