Downloads · 30 days
4
17% of all-time downloads
bn-regional-asr-300m
cykyy/bn-regional-asr-300m
bn-regional-asr-300m is a machine learning model from cykyy. Use it for the machine learning task on the model card, and read the license before you ship it in a product. The card lists the license as mit.
Fine-tuned version of omniASRCTC300Mv2 on RegSpeech12 Bengali regional speech dataset.
All-time downloads
23
Public
Repo size
1.3 GB
Likes
0
Public
Repo makeup
Click a slice to open those files.
.pt1.3 GB · 100%
At a glance
- License
- mit
- Access
- Public
- Created
- Jan 19, 2026
- Updated
- Jan 19, 2026
- SHA
- 7e7d87ee
Base models
Model card
From the Hugging Face model README
Bengali Regional ASR Model (300M)
Fine-tuned version of omniASR_CTC_300M_v2 on RegSpeech12 Bengali regional speech dataset.
Model Details
- Base Model: omniASR_CTC_300M_v2 (300M parameters)
- Tokenizer: omniASR_tokenizer_written_v2
- Training Data: RegSpeech12 (17,049 samples, ~80hr train, ~10hr valid)
- Training Steps: 10,000
- Epochs: 13.6
- Learning Rate: 5e-05
- Optimizer: AdamW (β₁=0.9, β₂=0.98, weight_decay=0.01)
- LR Scheduler: Tri-stage (10% warmup, 40% hold, 50% decay)
- Training Time: ~6.4 hours (single GPU)
- Final WER: 73.5% (on RegSpeech12 test set)
Usage
With OmniLingual ASR
- Create model card at
~/.config/fairseq2/assets/model/bn_regional.yaml:
name: bn_regional_CTC_300M_v2
model_family: wav2vec2_asr
model_arch: 300m_v2
checkpoint: /path/to/model.pt
tokenizer_ref: omniASR_tokenizer_written_v2
- Use with run_pipeline.py or ASRInferencePipeline
License
MIT
- License
- mit
- Languages
- bn
- Created
- Jan 19, 2026
- Updated
- Jan 19, 2026