techolise/akan-twi-speaker17-tts-v2

🇬🇭 Akoma Akan (Asante Twi) TTS V2 — Speaker 17 Master Voice

Production-grade Text-to-Speech system for Asante Twi (Akan), trained on the 15,560-sample multi-speaker speech corpus (ghanaopendata/twi-speech-text-multispeaker-16k) and specialized on the native female voice of Speaker 17 (Age 21).

🚀 Key Features

• Multi-Speaker Pronunciation Pretrained: Trained across 15,560 native Twi sentences to eliminate foreign accents and syllable miscuing.
• Twi Text Normalizer Integrated: Built-in support for Twi numbers, currency (GH₵), dates, and phonetic diacritics (ɛ, ɔ).
• Speaker 17 Master Voice: Warm, melodic, 21-year-old native female voice profile.
• High-Fidelity Diffusion: Powered by Continuous Flow Matching (CFM) DiT with Neural ZipEnhancer.

💻 Quick Python Usage

`python from voxcpm import VoxCPM from huggingface_hub import snapshot_download import soundfile as sf

1. Load Base & Speaker 17 V2 Model

base_dir = snapshot_download('ghananlpcommunity/ghana-tts-36k') spk17_dir = snapshot_download('techolise/akan-twi-speaker17-tts-v2')

model = VoxCPM(base_dir, enable_denoiser=True) anchor_wav = f'{spk17_dir}/speaker_17_dataset/speaker17_golden_anchor.wav' anchor_txt = 'Atoyerɛnkyɛm asi wɔ bea a wɔgu fangoo ɛgu ɛhyɛn mu.'

2. Synthesize ANY Twi sentence

text = 'Mema mo akye me nuanom nyinaa! Wo ho te sɛn? Akwaaba ba Ghana!' wav = model.generate( text=text, prompt_wav_path=anchor_wav, prompt_text=anchor_txt, cfg_value=2.4, inference_timesteps=26 )

sf.write('twi_output.wav', wav, 16000) `

🌐 Live Web Studio

• Web App: https://techolise-akan-twi-tts-web.hf.space
• Hugging Face Space: https://huggingface.co/spaces/techolise/akan-twi-tts-web