techiaith/kaldi-cy

Kaldi / Vosk Welsh Speech Recognition — 2607

A Kaldi chain (LF-MMI) acoustic model with WFST decoding, packaged for Vosk. It runs in real time on CPU, which makes it suited to on-device and streaming use where an end-to-end model is too costly.

Welsh transcription only. Output carries no punctuation or casing.

Evaluation results

WER / CER (lower is better) on the held-out techiaith/asr-benchmark-cy-en-2606 test sets, scored per-utterance against the same references for every model compared.

Mean across the Welsh test sets: WER 47.25, against 50.52 for the previous release (2503) — an improvement of 3.27 points.

Test set	WER	CER	WER (2503)
cymen-arfor/lleisiau-arfor (spontaneous, mixed dialects)	55.99	28.05	54.45
techiaith/banc-trawsgrifiadau-bangor (spontaneous, mixed dialects)	40.23	17.49	46.03
techiaith/commonvoice-23-0-cy (read)	45.52	15.92	51.08

Training data

Welsh speech corpora used to train this model:

Dataset	Split	Revision	Access
techiaith/banc-trawsgrifiadau-bangor	train	26.07	public
techiaith/corpws-clllc-wlga	clips	5d00294c31	public
techiaith/commonvoice_23_0_cy	train+dev[50%:]	main	private †
cymen-arfor/lleisiau-arfor	train	main	public
techiaith/commonvoice_vad_cy	train	main	public

† Common Voice data is not redistributed here. techiaith/commonvoice_23_0_cy (Common Voice 23.0 Welsh) is held in a private repository, in line with the Mozilla Foundation's terms for obtaining and sharing Common Voice data. It is listed above for provenance and reproducibility, not as a download.

The Welsh Common Voice corpus is freely available from Mozilla themselves: Mozilla Common Voice — Welsh datasets. Releases are distributed through the Mozilla Data Collective, and the datacollective package offers a programmatic download. With that in hand the rest of this pipeline is reproducible — every other dataset in the table is openly available on the Hugging Face Hub. Our thanks to Mozilla and to the Welsh speakers who contributed their voices.

Welsh transcription audio only, as materialised for training:

Source	Utterances	Hours
btb	50,969	58.9
wlga	29,222	53.4
lla	27,465	28.0
cv23	10,721	15.5
cv_vad	8,205	3.5
total	126,582	159.4

Training configuration

Parameter	Value
Recipe	local/chain/run_tdnn_260412.sh (chain / LF-MMI)
Acoustic model	TDNN-F, 12 factorised layers, 5.1M parameters
Tree leaves	3500
Epochs	5
Features	40-dim MFCC (hires) + 100-dim online i-vectors
SpecAugment	yes (spec-augment-layer in the xconfig)
Language model	3-gram, Witten-Bell discounting, trained on the training transcripts only
Lexicon	Geiriadur Ynganu Bangor pronunciations, with grapheme-to-phoneme fallback
Decoding vocabulary	62,392 words
Held-out dev WER	30.91% (BTB 32.60%, Common Voice 24.85%)

Usage

Install vosk, then unpack the bundle:

pip install vosk
huggingface-cli download techiaith/kaldi-cy model_cy.tar.gz --local-dir .
tar -xzf model_cy.tar.gz          # -> ./model/

import json
import wave

from vosk import KaldiRecognizer, Model

model = Model("model")

wf = wave.open("welsh_audio.wav", "rb")   # 16 kHz, mono, 16-bit PCM
rec = KaldiRecognizer(model, wf.getframerate())

results = []
while True:
    data = wf.readframes(4000)
    if len(data) == 0:
        break
    if rec.AcceptWaveform(data):
        results.append(json.loads(rec.Result())["text"])
results.append(json.loads(rec.FinalResult())["text"])

print(" ".join(t for t in results if t).strip())

Feed the audio in chunks and collect Result() each time AcceptWaveform returns True, as above. A single AcceptWaveform call followed by only FinalResult() discards everything recognised before the last detected end-of-utterance, which silently truncates anything longer than one utterance.

Limitations

• Welsh only. English audio is not a supported task.
• Closed vocabulary. WFST decoding can only emit words present in the lexicon (62,392 words), which is built from the training transcripts. Names and loanwords outside it cannot be produced at all — a hard floor on achievable WER.
• No punctuation or casing.
• The language model is a 3-gram trained on the training transcripts only, so the model is weaker on domains and registers unlike its training data.

Acknowledgements

Built with the Kaldi/Vosk recipe developed by Gweltaz Duval-Guennoc, Preben Vangberg and Sasha Wanasky (vosk-cymraeg).

Developed by Uned Technolegau Iaith, Prifysgol Bangor / Language Technologies Unit, Bangor University.

Funded by the Welsh Government.