Model card
Browse files
README.md
ADDED
|
@@ -0,0 +1,29 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: apache-2.0
|
| 3 |
+
language:
|
| 4 |
+
- tr
|
| 5 |
+
tags:
|
| 6 |
+
- onnx
|
| 7 |
+
- onnxruntime-web
|
| 8 |
+
- webgpu
|
| 9 |
+
- speech-recognition
|
| 10 |
+
- punctuation
|
| 11 |
+
- text-to-speech
|
| 12 |
+
---
|
| 13 |
+
|
| 14 |
+
# editor-models
|
| 15 |
+
|
| 16 |
+
Browser-ready files for [editor](https://github.com/batikankutluer/editor), a Turkish subtitle and
|
| 17 |
+
video editor that runs its models in the browser (onnxruntime-web, WebGPU). These are the files
|
| 18 |
+
that have no copy where their models are published; everything else (Whisper, Silero VAD, Antalia
|
| 19 |
+
Mini, the decision model) the editor fetches from the original repositories.
|
| 20 |
+
|
| 21 |
+
| Directory | What | Source and license |
|
| 22 |
+
| --- | --- | --- |
|
| 23 |
+
| `seda-v0.1/` | Seda Turkish streaming ASR, encoder rewritten for WebGPU (`encoder.webgpu.onnx`: MatMulInteger → DequantizeLinear + MatMul, batch fixed to 1, `pad_mask` input), `lodr.bin` (2gram.fst in a flat CSR layout), the decoder, joiner, RNN-LM and tokens unchanged | [atasoglu/seda-v0.1](https://huggingface.co/atasoglu/seda-v0.1), Apache-2.0; the LM files (`rnnlm`, `lodr`) CC BY-SA 4.0 (trained on Turkish Wikipedia) |
|
| 24 |
+
| `punct/` | Punctuation, true-casing and sentence boundaries, 47 languages, dynamic int8 (`punct.int8.onnx`), its SentencePiece vocabulary as JSON | [1-800-BAD-CODE/punct_cap_seg_47_language](https://huggingface.co/1-800-BAD-CODE/punct_cap_seg_47_language), see its model card for the license |
|
| 25 |
+
| `ema-lightning/` | EMA Lightning Turkish text-to-speech, ONNX graphs and the `normalizer-tr` wasm as served by ema-lightning-web | [canberkkkkkk/ema-lightning](https://huggingface.co/canberkkkkkk/ema-lightning), Apache-2.0; [normalizer-tr](https://github.com/erdemtuna/normalizer-tr), Apache-2.0 |
|
| 26 |
+
| `antalia-mini/` | Antalia-2 Mini Turkish text-to-speech (Patientdesk.ai), the "mixed" precision graphs and frontend data of its Space at commit `2c24b406`, with a `model.json` listing the file sizes | [cloud0day3/antalia-mini](https://huggingface.co/cloud0day3/antalia-mini), Apache-2.0 |
|
| 27 |
+
|
| 28 |
+
Each directory has a `model.json` (hyper-parameters and file sizes) the editor reads first. The
|
| 29 |
+
conversion script is `scripts/prepare_models.py` in the editor repository.
|