Add model card
Browse files
README.md
ADDED
|
@@ -0,0 +1,39 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: apache-2.0
|
| 3 |
+
base_model: MediaTek-Research/Breeze-ASR-26
|
| 4 |
+
language: [zh, nan]
|
| 5 |
+
pipeline_tag: automatic-speech-recognition
|
| 6 |
+
tags: [whisper, whisper.cpp, ggml, gguf, taiwanese-hokkien, taigi, macwhisper, superwhisper, edge, quantized, speech-recognition]
|
| 7 |
+
---
|
| 8 |
+
|
| 9 |
+
# Breeze-ASR-26 — GGML (whisper.cpp, Q4_0 / Q5_0)
|
| 10 |
+
|
| 11 |
+
The **smallest-footprint** build: peak RSS 1.85 GB (Q4_0), fits a 4 GB host, and
|
| 12 |
+
loads directly in **whisper.cpp desktop apps (MacWhisper Pro, superwhisper)** —
|
| 13 |
+
drag the .bin in, no code. Slower than CT2 (RTF 0.40) but the RAM floor is 1 GB lower.
|
| 14 |
+
|
| 15 |
+
(Files are whisper.cpp GGML format. whisper.cpp GGUF support is in progress upstream,
|
| 16 |
+
ggml-org/whisper.cpp#3316; GGUF builds will be added when the toolchain lands.)
|
| 17 |
+
|
| 18 |
+
> Part of the **Breeze-ASR-26 edge family** — the same MediaTek model in every runtime, pick by your constraint:
|
| 19 |
+
>
|
| 20 |
+
> | Repo | Runtime | RSS | RTF (CPU 4-thread) | Best for |
|
| 21 |
+
> |---|---|---|---|---|
|
| 22 |
+
> | [Breeze-ASR-26-ct2](https://huggingface.co/weemed/Breeze-ASR-26-ct2) | CTranslate2 / faster-whisper | ~2.9 GB | **0.21** | servers, 8 GB+ hosts, GPU |
|
| 23 |
+
> | [Breeze-ASR-26-GGML](https://huggingface.co/weemed/Breeze-ASR-26-GGML) | whisper.cpp / MacWhisper | **1.85 GB** | 0.40 | 4 GB hosts, desktop apps |
|
| 24 |
+
> | [Breeze-ASR-26-ONNX](https://huggingface.co/weemed/Breeze-ASR-26-ONNX) | sherpa-onnx / onnxruntime | — | 1.3 | Android / iOS / WASM |
|
| 25 |
+
>
|
| 26 |
+
> All Apache-2.0, derived from [MediaTek-Research/Breeze-ASR-26](https://huggingface.co/MediaTek-Research/Breeze-ASR-26). Measured on real multi-speaker Mandarin meeting audio. Mandarin does not regress; Taigi is transcribed as Mandarin meaning (not verbatim Taigi characters).
|
| 27 |
+
|
| 28 |
+
## Usage
|
| 29 |
+
|
| 30 |
+
```bash
|
| 31 |
+
# whisper.cpp
|
| 32 |
+
whisper-cli -m breeze-q5_0.bin -f meeting.wav -l zh -t 4
|
| 33 |
+
```
|
| 34 |
+
**MacWhisper Pro / superwhisper**: download `breeze-q5_0.bin`, drag it in as a custom model.
|
| 35 |
+
|
| 36 |
+
| Build | Disk | Peak RSS | RTF |
|
| 37 |
+
|---|---|---|---|
|
| 38 |
+
| Q4_0 | 848 MB | 1.85 GB | 0.40 |
|
| 39 |
+
| Q5_0 | 1.1 GB | 2.03 GB | 0.62 |
|