btsee/common-voice-26-mn

Common Voice 26.0 Mongolian (cleaned)

A quality-filtered, normalised subset of Mozilla Common Voice Corpus 26.0, Mongolian, prepared for training Mongolian (Khalkha Cyrillic) text-to-speech with oron-tts.

Built by oron-cleaner. Every threshold was calibrated on this corpus, and every number and column on this page is read from the shipped data rather than asserted.

from datasets import load_dataset
ds = load_dataset("btsee/common-voice-26-mn", split="train")
print(ds[0]["text"], ds[0]["audio"]["sampling_rate"])

What is in it

clips	15,092
audio	16.33 h
distinct sentences	5,635
identified speakers	378
format	parquet shards, audio inline as 24 kHz mono wav
upstream audio	up to 48 kHz MP3, variable (crowd-sourced)
median clip	3.77 s
median measured bandwidth	7,570 Hz
median DNSMOS OVR	3.75
median alignment score	0.882
median ASR CER	0.071

Splits

split	clips	hours	speakers
test	75	0.07	63
train	14515	15.74	252
validation	115	0.13	63
withheld	387	0.40	95

All rows ship in one file set; the split column says which split a clip belongs to, so you can filter without re-downloading.

withheld holds clips whose sentences were removed from train, so text used as an evaluation target is genuinely unseen. It is a text holdout, not a speaker holdout — the two are separate requirements, and intersecting them leaves too little data to be useful.

Gender

gender	clips	hours
female	6731	7.33
male	8194	8.82
unknown	167	0.18

gender_resolved is declared where the contributor gave it, propagated across a speaker's other clips where one of them declared it, and otherwise inferred from median F0. gender_source records which of the three applied, so inferred labels can be dropped if you do not trust them.

Columns

column	type	example
audio	audio	24 kHz mono wav, stored inline
accents	string	Халх
age	string	twenties
align_score	float64	0.902277702492957
asr_transcript	string	нөөц болон бүтээг түүний тусгайрлалтань остр
audio_path	string	wavs/cv_common_voice_mn_18971635.wav
bandwidth_hz	float64	7453.125
cer	float64	0.12149532710280374
client_id	string	53772662a3f484241607d97212027683dc7005888cfe
clip_id	string	cv_common_voice_mn_18971635
dnsmos_bak	float64	3.2829662174514005
dnsmos_ovr	float64	4.1021785665940245
dnsmos_p808	float64	3.0004064527902163
dnsmos_sig	float64	3.686748186747233
down_votes	int64	0
duration_s	float64	7.996
duration_tsv	float64	9.936
gender	string	male_masculine
gender_resolved	string	male
gender_source	string	f0
len_ratio	float64	1.0093457943925233
locale	string	mn
mean_f0_hz	float64	104.4660675955349
path	string	common_voice_mn_18971635.mp3
pitch_confidence	float64	0.29
segment	string
sentence	string	Нөөц болон бүтээгдэхүүний тусгаарлалт нь Ост
snr_db	float64	38.240480340962584
split	string	train
text	string	Нөөц болон бүтээгдэхүүний тусгаарлалт нь Ост
up_votes	int64	2
variant	string

What the derived columns mean

column	meaning
text	normalised transcript — the training target
clip_id	stable identifier, unique in this corpus
audio_path	path in the source corpus layout, kept for traceability
gender_resolved / gender_source	see above
snr_db	speech-region RMS against true non-speech RMS
bandwidth_hz	measured lowpass shelf, not the container sample rate
dnsmos_sig / dnsmos_bak / dnsmos_ovr / dnsmos_p808	learned perceptual quality
align_score	MMS-FA forced-alignment confidence against text
cer / len_ratio / asr_transcript	ASR agreement with text
mean_f0_hz / pitch_confidence	pitch, used for gender inference
duration_s	duration of the shipped audio, after edge trimming
split	train, validation, test or withheld

How it was cleaned

Each clip passes every gate below or is dropped. Gates run in order, and a clip stops at its first failure.

1. Duration — outside 1–20 s is dropped.
2. Clipping and DC offset — consecutive full-scale samples, or a DC bias.
3. Silero VAD — a clip that is mostly silence is a failed recording.
4. SNR — speech-region RMS against true non-speech RMS on the untrimmed signal. When a clip is spoken end to end there is no noise floor to measure; that is recorded as unmeasurable rather than treated as a failure, and the background judgement is left to DNSMOS-BAK.
5. Bandwidth — the real lowpass shelf, which is what an MP3 encoder or a 16 kHz source leaves behind.
6. DNSMOS (SIG / BAK / OVR) — learned perceptual quality.
7. MMS-FA forced alignment against the given transcript. This is the primary transcript gate: a low score means the audio does not contain those words.
8. ASR CER as a secondary check.

Every measurement is kept as a column, so you can re-filter to your own thresholds instead of trusting ours.

Text is normalised with oron_tts.text.MongolianNormalizer: numeral expansion that refuses rather than guesses case suffixes, small-caps folding (ЭЗЭНий becomes Эзэний), and abbreviation and unit expansion. The normalised string is simultaneously the published text, the CER reference and the training target, so a wrong expansion would be published, scored against itself, and learned.

provenance.json records the pinned model and dataset revisions, package versions, a fingerprint of the normaliser source, and a content hash.

=== Corpus summary ===

split          clips    hours  speakers   male h  female h
train         14,515     15.7       252      8.5       7.1
validation       115      0.1        63      0.1       0.1
test              75      0.1        63      0.0       0.0
withheld         387      0.4        95      0.2       0.2

Text diversity:
  clips                 15,092
  distinct sentences    5,635  (2.68x repetition)
  letters covered       34/35
  letter bigrams        734 distinct
  NEVER APPEARS         щ

CER by source (the gate's recogniser trained on Common Voice):
  source         clips   median    mean
  cv             15092    0.071   0.079
  A markedly lower median for cv is contamination, not quality.

total: 16.3 h   male 8.8 h   female 7.3 h

Acceptance criteria:
  total >= 25 h            FAIL  (16.3 h)
  male  >= 5 h             PASS  (8.8 h)
  male speakers >= 3       PASS  (270)

Using it for oron-tts finetuning

python scripts/build_f5_dataset.py --corpus <corpus-dir> --name oron_mn --splits train
python scripts/preflight.py --data data/oron_mn_pinyin --target-updates 30000

manifest.jsonl ships alongside the parquet in the original corpus layout, which is what build_f5_dataset.py expects.

Considerations

• Licence: CC0-1.0. Contributors dedicated their recordings to the public domain under CC0 for speech research. They did not consent to individual voice cloning; see Considerations below.
• Voice cloning. This corpus is suitable for training zero-shot TTS, which can clone a voice from seconds of audio. Contributors consented to speech research, not to having their individual voice reproduced. Do not synthesise a named or identifiable person's voice without that person's consent.
• Coverage. One language variety, read speech. Accent, dialect and spontaneous-speech coverage is untested.
• Inferred gender labels. Where gender_source is f0, the label is a guess from pitch, not a self-report. Filter on gender_source if that matters.
• The gates encode judgements. Thresholds were calibrated on this data; a different application may want different ones.

Citation

@misc{orontts_cv_mn,
  title  = {Common Voice 26.0 Mongolian (cleaned)},
  author = {Badral, Battseren},
  year   = {2026},
  url    = {https://huggingface.co/datasets/btsee/common-voice-26-mn}
}

Please also cite the upstream corpus: Mozilla Common Voice Corpus 26.0, Mongolian.

Used by

btsee/oron-tts — Mongolian text-to-speech, trained on this corpus.