GLM-5.3 is live. Run Z.ai's latest model on Friendli Model APIs. Try it today ➜
SinhaSpeech
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Run this model inference with full control and performance in your environment.
Model Details
Model Provider
SinhaSpeech
Model Tree
openai/whisper-small
Input Modalities
Output Modalities
Supported Functionality
stratified_v6 — 123,205 train / 15,763 validation / 15,860 test rows, speaker-disjoint. v6 canonicalizes ~2,954 word instances (letter-level spelling, ZWJ/rakaransaya conjuncts, word-level register pairs, and English typos in code-switched references) relative to stratified_v5.| Split | WER | CER |
|---|---|---|
| Validation (final, epoch 5) | 20.15% | 6.23% |
| Test (full, 15,860 rows) | 17.17% | 4.71% |
For comparison, the otherwise-identical model trained on stratified_v5 (same hyperparameters, no data canonicalization) scored 17.15% WER / 4.62% CER on the same test set — essentially a statistical tie. The v6 data cleanup did not produce a measurable test-set improvement at this configuration, despite winning validation WER on 4 of 5 epochs.