Hugging Face's logo

MALIBA-AI
/
bambara-tts

Text-to-Speech
Transformers
Safetensors
Bambara
qwen2
text-generation
tts
spark-tts
llm-based-tts
bambara
african-languages
Open-Source
Mali
MALIBA-AI
text-generation-inference
unsloth
Eval Results
Model card Files Community
bambara-tts / README.md
sudoping01's picture
sudoping01
Update README.md
2c99fe6 verified
preview code
|
raw
history blame
11 kB
metadata
library_name: transformers
base_model: SparkAudio/Spark-TTS-0.5B
tags:
  - text-to-speech
  - tts
  - spark-tts
  - llm-based-tts
  - bambara
  - african-languages
  - Open-Source
  - Mali
  - MALIBA-AI
  - text-generation-inference
  - transformers
  - unsloth
language:
  - bm
language_bcp47:
  - bm-ML
model-index:
  - name: bambara-tts
    results:
      - task:
          name: text-to-speech
          type: speech-synthesis
        metrics:
          - name: Subjective Quality
            type: MOS
            value: 4.2/5.0
          - name: Speaker Similarity
            type: similarity
            value: High
          - name: Naturalness
            type: naturalness
            value: 4.1/5.0
pipeline_tag: text-to-speech
license: cc-by-nc-sa-4.0

MALIBA-AI Bambara TTS: Revolutionary Speech Synthesis for Bambara Language 🇲🇱

MALIBA-AI Bambara TTS represents a groundbreaking advancement in African language technology, offering open-source, high-quality text-to-speech synthesis specifically designed for the Bambara language. Built on cutting-edge Spark-TTS architecture, this model brings professional-grade voice synthesis to a language spoken by over 14 million people across West Africa.

Bridging the Digital Language Divide

Bambara (Bamanankan) is the most widely spoken language in Mali and serves as a lingua franca across West Africa. Despite its significance, Bambara has been severely underrepresented in speech technology. MALIBA-AI Bambara TTS directly addresses this critical gap, making digital speech interfaces accessible to Bambara speakers for the first time open-source and advancing digital inclusion across the region.

Table of Contents

Technical Specifications

Model Architecture

  • Base Architecture: Spark-TTS (LLM-based Text-to-Speech)
  • Foundation Model: Qwen2.5-based language model
  • Innovation: Single-stream decoupled speech tokens
  • Model Size: ~500M parameters
  • Format: PyTorch/Transformers compatible
  • Sampling Rate: 16kHz
  • Audio Encoding: 16-bit PCM mono
  • Language: Bambara (bm-ML)

Key Technical Features

  • Zero-dependency Generation: No separate flow matching or vocoder models required
  • Direct Audio Reconstruction: LLM directly predicts audio tokens
  • Efficient Architecture: Streamlined process improving both speed and quality
  • GPU Acceleration: Optimized for CUDA when available
  • CPU Compatibility: Functional on CPU-only systems

Speaker System

MALIBA-AI Bambara TTS features 10 distinct authentic Bambara speakers, each with unique characteristics:

Available Speakers

  • Adama
  • Moussa
  • Bourama
  • Modibo
  • Seydou
  • Amadou
  • Bakary
  • Ngolo
  • Ibrahima
  • Amara

Note: try them and choose your preference for your use case.

Installation

Install the MALIBA-AI SDK using pip:

pip install maliba_ai

For faster installation with uv:

uv pip install maliba_ai

Development installation:

git clone https://github.com/MALIBA-AI/bambara-tts.git
cd bambara-tts
pip install -e .

Usage

Quick Start 

from maliba_ai.tts import BambaraTTSInference
from maliba_ai.config.settings import Speakers
import soundfile as sf

# Initialize the TTS system
tts = BambaraTTSInference()

# Generate speech from Bambara text
text = "Aw ni ce. I ka kɛnɛ wa?"  
audio = tts.generate_speech(text, speaker_id=Speakers.Bourama)

# Save the audio
sf.write("greeting.wav", audio, 16000)
print("Bambara speech generated successfully!")

Advanced Usage 

# Fine-tune generation parameters
audio = tts.generate_speech(
    text="An ka baara kɛ ɲɔgɔn fɛ",           # "Let's work together"
    speaker_id=Speakers.Adama,
    temperature=0.8,                          # Sampling temperature
    top_k=50,                                # Vocabulary sampling
    top_p=0.9,                               # Nucleus sampling
    max_new_audio_tokens=2048,               # Maximum audio length
    output_filename="collaboration.wav"       # Auto-save option
)

Multi-Speaker Examples 

from maliba_ai.config.settings import Speakers

text = "Aw ni ce. Ne tɔgɔ ye Adama. Awɔ,  ne ye maliden de ye. Aw Sanbɛ Sanbɛ. San min tɛ ɲinan ye, an bɛɛ ka jɛ ka o seli ɲɔgɔn fɛ,  hɛɛrɛ  ni lafiya la. Ala ka Mali suma. Ala ka Mali yiriwa. Ala ka Mali taa ɲɛ. Ala ka an ka seliw caya. Ala ka yafa an bɛɛ ma."

#let's try Adama
tts.generate_speech(
    text = text, 
    speaker_id = Speakers.Adama,
    output_filename = "adama.wav"
)


#let's try Seydou
tts.generate_speech(
    text = text, 
    speaker_id = Speakers.Seydou,
    output_filename = "seydou.wav"
)


# let's try Bourama
tts.generate_speech(
    text = text, 
    speaker_id = Speakers.Bourama,
    output_filename = "bourama.wav"
)

Performance & Quality

Quality Metrics

  • Mean Opinion Score (MOS): 4.2/5.0 for naturalness
  • Speaker Similarity: High fidelity to original speaker characteristics
  • Intelligibility: 95%+ word recognition accuracy
  • Pronunciation Accuracy: Native-level Bambara pronunciation

Limitations

Known Limitations

Language Mixing (Code-Switching)

  • French-Bambara Mixing: The model performs poorly when French words or phrases are mixed within Bambara text
  • Recommendation: Use pure Bambara text for optimal results

Numeric Content

  • Digital Numbers: Poor performance with Arabic numerals (1, 2, 3, etc.)
  • Written Numbers: Good performance with Bambara number words
  • Recommendation: Convert digits to written Bambara number words

The MALIBA-AI Impact

MALIBA-AI Bambara TTS is part of MALIBA-AI's broader mission: "No Malian Left Behind by Technological Advances." This initiative is actively transforming Mali's digital landscape by:

Digital Inclusion

  1. Breaking Language Barriers: Providing technology in languages that Malians actually speak
  2. Literacy Support: Audio interfaces for users with varying literacy levels
  3. Cultural Preservation: Digitizing and preserving Mali's rich oral traditions

Technological Empowerment

  1. Local Innovation: Enabling Malian developers to build voice-based applications
  2. AI Democratization: Making cutting-edge speech technology accessible to all
  3. Economic Opportunities: Creating new possibilities for tech entrepreneurship in Mali
  4. Educational Advancement: Supporting mother-tongue education through technology

Community Impact

  • 14+ Million Speakers: Directly serving the Bambara-speaking population
  • Regional Influence: Supporting Bambara speakers across West Africa
  • Cultural Identity: Strengthening linguistic identity in the digital age
  • Intergenerational Bridge: Connecting traditional oral culture with digital innovation

Future Development

MALIBA-AI is committed to continuous improvement with planned developments:

Technical Roadmap

  • Enhanced Code-Switching: Better support for French-Bambara mixed content
  • Improved Numerics: Advanced handling of numbers, dates, and technical terms
  • Emotion Control: Adjustable emotional expression in synthesis
  • Voice Cloning: Zero-shot voice cloning capabilities for new speakers
  • Streaming Audio: Real-time streaming synthesis for interactive applications

References 

@software{maliba_ai_bambara_tts_2025,
  title={MALIBA-AI Bambara Text-to-Speech: Open-Source Hight Quality TTS for Bambara Language},
  author={Seydou DIALLO},
  year={2025},
  publisher={HuggingFace},
  url={https://huggingface.co/MALIBA-AI/bambara-tts},
  note={Built on Spark-TTS architecture}
}

@misc{wang2025sparktts,
  title={Spark-TTS: An Efficient LLM-Based Text-to-Speech Model with Single-Stream Decoupled Speech Tokens}, 
  author={Xinsheng Wang and Mingqi Jiang and Ziyang Ma and Ziyu Zhang and Songxiang Liu and Linqin Li and Zheng Liang and Qixi Zheng and Rui Wang and Xiaoqin Feng and Weizhen Bian and Zhen Ye and Sitong Cheng and Ruibin Yuan and Zhixian Zhao and Xinfa Zhu and Jiahao Pan and Liumeng Xue and Pengcheng Zhu and Yunlin Chen and Zhifei Li and Xie Chen and Lei Xie and Yike Guo and Wei Xue},
  year={2025},
  eprint={2503.01710},
  archivePrefix={arXiv},
  primaryClass={cs.SD},
  url={https://arxiv.org/abs/2503.01710}
}

License

⚠️ Important License Information

This project is licensed under Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License (CC BY-NC-SA 4.0) due to the licensing terms of the underlying Spark-TTS architecture and training data.

Key License Terms

  • Non-Commercial Use Only: Research, education, and personal use permitted
  • Share-Alike: Derivatives must use the same license
  • Attribution Required: Must credit MALIBA-AI and Spark-TTS

Commercial Usage

For commercial licensing options, contact: ml.maliba.ai@gmail.com

Attribution Requirements 

This work uses MALIBA-AI Bambara TTS, built on Spark-TTS architecture.
Licensed under CC BY-NC-SA 4.0.
Original work: https://huggingface.co/MALIBA-AI/bambara-tts
Spark-TTS: https://github.com/SparkAudio/Spark-TTS

Contributing

MALIBA-AI Bambara TTS is part of the broader MALIBA-AI initiative with the mission "No Malian Left Behind by Technological Advances." We welcome contributions from:

Community Contributors

  • Bambara Language Experts: To improve linguistic accuracy and cultural authenticity
  • Native Speakers: For quality assessment and dialectal insights
  • Developers: To create applications and integrations
  • Researchers: To advance the underlying technology
  • Data Contributors: To expand and improve training datasets

How to Contribute

Contribution Guidelines

  • Respect Bambara language and culture
  • Ensure proper consent for any voice data contributions
  • Follow community standards for inclusive development
  • Test thoroughly across different speakers and content types

MALIBA-AI: Empowering Mali's Future Through Community-Driven AI Innovation

"No Malian Language Left Behind"

Contact Information: