Views
No views yet
| Base model | F5-TTS v1 Base (flow-matching DiT) |
| Parameters | 335M |
| Training data | ~13.4 hours (BibleTTS + SLR86 + WAXAL) |
| Steps | 150,000 |
| GPUs | 2× A100-40GB |
| Effective batch | 38,400 frames/step |
| Learning rate | 7.5e-5 |
| Tokenizer | Character-level (no pinyin) |
| Vocab | 2,562 chars (base + Yorùbá diacritics) |
1import f5_tts.model.utils as f5_utils
2# CRITICAL: Must bypass pinyin conversion BEFORE any other F5-TTS import
3f5_utils.convert_char_to_pinyin = lambda texts, polyphone=True: texts
4
5from f5_tts.api import F5TTS
6
7f5tts = F5TTS(
8 model="F5TTS_v1_Base",
9 ckpt_file="model_150000.pt",
10 vocab_file="vocab.txt",
11 device="cuda",
12)
13
14# ref_file: 5-10s WAV of any voice (add ~1s trailing silence for best results)
15# ref_text: what the reference says (Yorùbá or English)
16# gen_text: Yorùbá text with full diacritics
17wav, sr, _ = f5tts.infer(
18 ref_file="reference.wav",
19 ref_text="text spoken in reference",
20 gen_text="ẹ kú àárọ̀, báwo ni àwọn ọmọ yín ṣe wà?",
21 speed=1.0,
22 nfe_step=16,
23 file_wave="output.wav",
24)[breath], [snap]) from reference text if present.| Source | Samples | Hours | Notes |
|---|---|---|---|
| BibleTTS Yorùbá | 7,560 | ~8h | Studio quality, single speaker |
| SLR86 | 3,583 | ~3h | Crowdsourced, male + female |
| WAXAL TTS | 1,492 | ~3h | Diacritics restored via Gemini |
1@misc{naijaml-f5tts-yoruba-2026,
2 title={F5-TTS Yorùbá: First Zero-Shot Voice Cloning TTS for Yorùbá},
3 author={NaijaML},
4 year={2026},
5 url={https://huggingface.co/naijaml/f5-tts-yoruba}
6}