Downloads · 30 days
9
35% of all-time downloads
tamil-stt-cnn-bigru-ctc
Manojakshai/tamil-stt-cnn-bigru-ctc
tamil-stt-cnn-bigru-ctc is a machine learning model from Manojakshai. Use it for the machine learning task on the model card, and read the license before you ship it in a product.
Custom Tamil Speech-to-Text model trained from scratch using PyTorch.
All-time downloads
26
Public
Repo size
45.4 MB
Likes
1
Public
Repo makeup
Click a slice to open those files.
.pt45.4 MB · 100%
At a glance
- Model type
- simple_stt
- Access
- Public
- Created
- Sep 16, 2026
- Updated
- Sep 16, 2026
- SHA
- 38a8d401
Model card
From the Hugging Face model README
Tamil STT — CNN + BiGRU + CTC
Custom Tamil Speech-to-Text model trained from scratch using PyTorch.
Architecture
Audio → Log-Mel Spectrogram → CNN → 3-layer Bidirectional GRU → Linear Layer → CTC → Tamil Transcript
Model Details
- Sample rate: 16 kHz
- Mel bins: 80
- FFT size: 400
- Hop length: 160
- RNN: 3-layer Bidirectional GRU
- Hidden size: 256
- Output: Character-level CTC
- Vocabulary size: 48
Experiment 2 Results
- Best checkpoint: best_cer.pt
- Best epoch: 82
- Validation CER: 21.13%
- Best validation loss: 1.0641
- Training approach: stronger frequency/time masking augmentation
Files
best_cer.pt— trained model checkpointconfig.json— model configurationvocab.json— character vocabulary
This repository contains the trained model checkpoint and supporting metadata.
- Type
- simple_stt
- Created
- Sep 16, 2026
- Updated
- Sep 16, 2026