This is the highest-quality open Amharic TTS model available today. It generates natural, expressive speech from text and can clone any speaker's voice from a 10-second audio sample.
Most "multilingual" TTS models (MMS, XTTS) produce Amharic that sounds robotic or mispronounces ejective consonants (ጠ, ጰ, ጸ, ፀ, ቸ, ጨ). This model was trained exclusively on Amharic audio and preserves:
Give it 10 seconds of any Amharic speaker and it will synthesize new sentences in that voice. Tested on:
1# Cell 1: Install
2!pip install -q omnivoice soundfile
3
4# Cell 2: Load model
5import torch
6from omnivoice import OmniVoice, OmniVoiceGenerationConfig
7
8model = OmniVoice.from_pretrained(
9 "african-low-resource/omnivoice-amharic",
10 device_map="cuda:0",
11 dtype=torch.float16,
12)
13
14# Cell 3: Generate speech
15text = "ሰላም፣ እንኳን ደህና መጣችሁ። ይህ የአማርኛ ንግግር ሙከራ ነው።"
16audio = model.generate(
17 text=text,
18 language="Amharic",
19 generation_config=OmniVoiceGenerationConfig(num_step=32, guidance_scale=2.0),
20)
21
22import soundfile as sf
23sf.write("output.wav", audio[0], 24000)
24print("✅ Saved to output.wav")
1# Upload a 10-second reference WAV
2prompt = model.create_voice_clone_prompt(ref_audio="speaker.wav", ref_text=None)
3
4audio = model.generate(
5 text="ዛሬ ቀን ጥሩ ነው።",
6 language="Amharic",
7 voice_clone_prompt=prompt,
8 generation_config=OmniVoiceGenerationConfig(num_step=32, guidance_scale=2.0),
9)
10sf.write("cloned.wav", audio[0], 24000)
We evaluate on a held-out test set (10% of combined data, never seen in training).
1@software{omnivoice_amharic_2026,
2 author = {demeleww and Voices For All},
3 title = {OmniVoice Amharic: Open Voice AI for 60M Speakers},
4 year = {2026},
5 url = {https://huggingface.co/african-low-resource/omnivoice-amharic},
6 license = {Apache-2.0}
7}
1@article{omnivoice2026,
2 title={OmniVoice: High-Quality Voice Cloning TTS for 600+ Languages},
3 journal={arXiv preprint arXiv:2604.00688},
4 year={2026}
5}
