MSA F5-TTS Arabic

Fine-tuned F5-TTS on a professional Arabic MSA voice, built by Yahya (NightPrince).

Field Value
Architecture F5-TTS v1 Base (Flow Matching + DiT)
Base model SWivid/F5-TTS
Training data 439 clips / ~70 min
Sample rate 24 kHz
Language Arabic MSA + full Tashkeel
Final loss 0.367 (20 epochs)
Author Yahya (NightPrince)

Usage

from f5_tts.api import F5TTS
from huggingface_hub import hf_hub_download

ckpt  = hf_hub_download("NightPrince/MSA-F5-TTS-Arabic", "model.pt")
vocab = hf_hub_download("NightPrince/MSA-F5-TTS-Arabic", "vocab.txt")

tts = F5TTS(model="F5TTS_v1_Base", ckpt_file=ckpt, vocab_file=vocab, device="cuda")

wav, sr, _ = tts.infer(
    ref_file="reference.wav",
    ref_text="ุงู„ู†ุต ุงู„ู…ุฑุฌุนูŠ",
    gen_text="ุจูุณู’ู…ู ุงู„ู„ูŽู‘ู‡ู ุงู„ุฑูŽู‘ุญู’ู…ูฐู†ู ุงู„ุฑูŽู‘ุญููŠู…ู",
    nfe_step=32
)
Downloads last month
44
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support

Model tree for NightPrince/MSA-F5-TTS-Arabic

Base model

SWivid/F5-TTS
Finetuned
(82)
this model