Fine-tuned checkpoint of OmniLingual ASR CTC 300M_v2 on a mixture of FLEURS Chichewa/Nyanja (2,653 examples, ~10.87 hours) and the Chichewa Trigrams speech-text parallel corpus (39,157 examples, ~16.31 hours — 150% of FLEURS train hours). Training used the fairseq2 wav2vec2 ASR recipe for 5,000 steps with the encoder frozen for the first 1,000 steps. Tokenizer: omniASR_tokenizer_written_v2.

Results

Dataset	WER (%)	CER (%)
FLEURS Chichewa test	36.56	7.95
Zambezi Voice Nyanja test	60.42	13.37

This model achieves the best Zambezi Voice WER (60.42%) across all trigram mixing ratios tested, outperforming the zero-shot omniASR CTC 3B baseline (62.10%) by 1.68 pp. For balanced performance across both FLEURS and Zambezi, the 1:1 ratio model (ai4good-labyrinth/omniASR-CTC-300M-v2-chichewa-mixed-1p0) is preferred (35.27% FLEURS / 60.79% Zambezi).

Training details

Base model: omniASR_CTC_300M_v2
Tokenizer: omniASR_tokenizer_written_v2
Training data: FLEURS Chichewa train (2,653 examples, 10.87 h) + Chichewa Trigrams (39,157 examples, 16.31 h)
Steps: 5,000 (best checkpoint at step 3,500 by FLEURS dev WER)
Dev WER at best checkpoint: 37.93% (FLEURS dev, 305 examples)
Learning rate: 1e-5
Encoder frozen: first 1,000 steps
Gradient accumulation: 4 batches
Mixed precision: bfloat16
Framework: fairseq2

Checkpoint format

The checkpoint is stored in fairseq2's sharded format. checkpoint/model/pp_00/tp_00/sdp_00.pt contains the full model state (training used a single GPU, so there is only one shard).

To load the model, use the OmniLingual ASR inference pipeline with the omniASR_CTC_300M_v2 model card and point the checkpoint path at sdp_00.pt.

ai4good-labyrinth/omniASR-CTC-300M-v2-chichewa-mixed-1p5

omniASR CTC 300M_v2 — fine-tuned on Chichewa/Nyanja (FLEURS + trigram 1.5:1)

Results

Training details

Checkpoint format

ai4good-labyrinth/
omniASR-CTC-300M-v2-chichewa-mixed-1p5