Refresh README — uniform WindyWord template with WER tier + dialect notes
Browse files
README.md
ADDED
|
@@ -0,0 +1,52 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: apache-2.0
|
| 3 |
+
tags:
|
| 4 |
+
- automatic-speech-recognition
|
| 5 |
+
- whisper
|
| 6 |
+
- windyword
|
| 7 |
+
- hindi
|
| 8 |
+
- hindi
|
| 9 |
+
library_name: transformers
|
| 10 |
+
pipeline_tag: automatic-speech-recognition
|
| 11 |
+
language:
|
| 12 |
+
- hi
|
| 13 |
+
---
|
| 14 |
+
|
| 15 |
+
# WindyWord.ai STT — Hindi Lingua (CPU INT8 (CTranslate2))
|
| 16 |
+
|
| 17 |
+
**Transcribes Hindi speech (Indo-European > Indo-Iranian > Indo-Aryan).**
|
| 18 |
+
|
| 19 |
+
> **Note:** Outputs Hindi audio as **Latin-script Hinglish, NOT Devanagari**. FLEURS-Devanagari WER ≈100% is a script mismatch, not a quality failure. Useful for code-switched / chat / SMS contexts. For Devanagari output, use a separate model (not yet shipped).
|
| 20 |
+
|
| 21 |
+
## Quality
|
| 22 |
+
|
| 23 |
+
- **FLEURS WER:** 102.5% (50-sample audit)
|
| 24 |
+
- **CER:** 0.999
|
| 25 |
+
- **Tier:** UNUSABLE-GAP ⭐
|
| 26 |
+
- **Source:** WindyWord Grand Rounds v2 audit (50-sample FLEURS)
|
| 27 |
+
|
| 28 |
+
## About this variant
|
| 29 |
+
|
| 30 |
+
This is the **ct2-int8** deployment format of our Hindi Lingua STT model. Load it via the `ct2-int8/` subfolder.
|
| 31 |
+
|
| 32 |
+
Part of the [WindyWord.ai](https://windyword.ai) STT fleet — covering 35+ languages that commercial speech-to-text APIs underserve, with proper dialect / script disclosures where they matter.
|
| 33 |
+
|
| 34 |
+
## Usage
|
| 35 |
+
|
| 36 |
+
```python
|
| 37 |
+
from transformers import WhisperForConditionalGeneration, WhisperProcessor
|
| 38 |
+
processor = WhisperProcessor.from_pretrained("WindyWord/listen-windy-lingua-hindi-ct2", subfolder="ct2-int8")
|
| 39 |
+
model = WhisperForConditionalGeneration.from_pretrained("WindyWord/listen-windy-lingua-hindi-ct2", subfolder="ct2-int8")
|
| 40 |
+
```
|
| 41 |
+
|
| 42 |
+
## Commercial Use
|
| 43 |
+
|
| 44 |
+
Visit [windyword.ai](https://windyword.ai) for apps and API access.
|
| 45 |
+
|
| 46 |
+
---
|
| 47 |
+
|
| 48 |
+
## Provenance & License
|
| 49 |
+
|
| 50 |
+
Weights derived from upstream community Whisper fine-tunes (see individual model card for exact lineage). Redistributed under Apache-2.0 (inherited).
|
| 51 |
+
|
| 52 |
+
*Certified by Opus 4.6 Opus-Claw (Dr. C) on Veron-1 (RTX 5090, Mt Pleasant SC).*
|