You can deploy this model and start earning money today!

btsee/
oron-tts

Name: oron-tts
Availability: InStock

Views

No views yet

OronTTS — Mongolian text to speech

Speaks Mongolian (Khalkha, Cyrillic) in two fixed voices, one male and one female. A finetune of F5-TTS on cleaned Mongolian speech from three public corpora.

Listen

Male:

Female:

Install

bash

pip install git+https://github.com/SWivid/F5-TTS.git
pip install git+https://github.com/btseee/oron-tts.git

Use

python

import soundfile as sf
from huggingface_hub import hf_hub_download
from f5_tts.api import F5TTS
from oron_tts.text import MongolianNormalizer

VOICE = "female"          # or "male"

ckpt  = hf_hub_download("btsee/oron-tts", "model.safetensors")
vocab = hf_hub_download("btsee/oron-tts", "vocab.txt")
ref   = hf_hub_download("btsee/oron-tts", f"voices/{VOICE}.wav")
rtxt  = hf_hub_download("btsee/oron-tts", f"voices/{VOICE}.txt")

tts = F5TTS(model="F5TTS_v1_Base", ckpt_file=ckpt, vocab_file=vocab, use_ema=False)
wav, sr, _ = tts.infer(
    ref_file=ref,
    ref_text=open(rtxt, encoding="utf-8").read().strip(),
    gen_text=MongolianNormalizer().normalize("Сайн байна уу. Өнөөдөр цаг агаар сайхан байна.",
                                             strict=True),
    nfe_step=32, cfg_strength=2.0, sway_sampling_coef=-1.0, seed=0)

sf.write("out.wav", wav, sr)

Two things that break it silently

Keep use_ema=False. The EMA weights synthesise fluent non-words, an order of magnitude worse by CER than the raw tensors, while sounding like confident speech -- so you will not hear the mistake.

Normalise the text. Anything outside the vocabulary is read as a space, because unknown ids map to index 0 and index 0 is the space token. Digits, Latin letters and punctuation all need MongolianNormalizer.

Numbers

Measured on the shipped voices/ prompts, over held-out sentences never used to select anything: n=201 per voice. Micro-CER and mean UTMOS, 95% bootstrap intervals.

	male	female
CER	0.0508 [0.0466–0.0552]	0.0630 [0.0585–0.0676]
UTMOS	2.95 [2.91–2.99]	2.95 [2.92–2.98]
speaker similarity to own prompt	0.741	0.717

The voices score 0.105 against each other; here real same-speaker pairs score 0.540–0.833, different speakers 0.034–0.503.

Per-checkpoint numbers in eval.json, curves in TensorBoard.

Limits

The output is wideband, roughly 8 kHz -- not full-band: the corpus pipeline decoded to 16 kHz before measuring. Not a limit of Mongolian audio; fixed after this model.
The CER above is optimistic: its scorer is fine-tuned on Common Voice, which is most of this model's training data, so read every CER against the ground-truth human floor, never against zero.
No listening test has been run -- UTMOS is a proxy trained on English and Japanese MOS data, never validated for Mongolian.
There is no watermarking; audio from this model cannot be detected as synthetic by any downstream tool.
Consent: Common Voice contributors dedicated their recordings CC0 for speech research and did not consent to having their individual voice cloned.
The normaliser refuses numeral case suffixes it cannot expand without guessing, so a little input is rejected rather than mispronounced.

Licence

CC-BY-4.0. Attribution is required because FLEURS is CC-BY-4.0. It does not train on WorldSpeech, so it carries no non-commercial restriction.

btsee/oron-tts