travis-moore/sema-pidgin-v0

Sema Pidgin v0

Sema's first Pidgin voice: a cloned voice that speaks Nigerian Pidgin and English. Give it a few seconds of reference audio and it speaks your lines in that voice.

Built under Sema by Angelo Asante.

The Sema Pidgin line aims at one voice with three legs — standard English, African-accented English, and Pidgin. This first version trains the Pidgin leg.

Hear it

Reference-cloned voices.

Clip	What it is
	Nigerian Pidgin, Nigerian voice: "Wetin dey happen? I dey come your side this evening..."
	Nigerian Pidgin, Nigerian voice: "No wahala, I don send the money..."
	Ghanaian Pidgin, author's voice: "Chale, wassup? I dey go chop bar..."
	English, author's voice
	A held-out sentence the model never trained on

How we built it

• Several hours of native Nigerian Pidgin speech, many speakers, short clips.
• We fine-tuned only the pronunciation stage (accent, rhythm and pronunciation), over 5 passes of the data, and left the voice encoder and vocoder frozen — so the model keeps clean zero-shot voice cloning and clean English while speaking Pidgin far more naturally than before.

The shipped weights are the point on the curve where the voice reads most naturally across general Pidgin, not the very last step.

Use

from huggingface_hub import snapshot_download
from chatterbox.tts import ChatterboxTTS
import torchaudio

d = snapshot_download("travis-moore/sema-pidgin-v0", token=HF_TOKEN)
model = ChatterboxTTS.from_local(d, device="cuda")
wav = model.generate(
    "No wahala, I don send the money. E go land before you reach house.",
    audio_prompt_path="reference.wav",   # any 7-10 s clip of the target voice
    exaggeration=0.5, cfg_weight=0.5,
)
torchaudio.save("out.wav", wav, model.sr)

The folder is a self-contained model directory; it loads with ChatterboxTTS.from_local.

Limitations

• Trained on Nigerian Pidgin, so Ghanaian and Cameroonian Pidgin come through by transfer and are lighter. A Nigerian voice reference gives the strongest Pidgin.
• Numbers, acronyms and dates are a known soft spot.
• Outputs carry an inaudible watermark.
• Clone only voices you have the right to use.

The combined Twi + Pidgin voice that follows this line is Sema Sauti.

Safety

This model clones a voice from a few seconds of audio. Use it only on voices you have the right to use. Impersonating a real person without their consent, deception and fraud are against the intent of this release; disclose synthetic audio where a listener could reasonably be misled.

---

Fine-tuned from ResembleAI/chatterbox (MIT).