termux-stt 1.2.15 → 1.3.1
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/README.md +30 -2
- package/README.pypi.md +8 -2
- package/package.json +1 -1
package/README.md
CHANGED
|
@@ -19,6 +19,22 @@ Integrates Whisper.cpp, Vosk, and Sherpa-ONNX with a closed-form pure-Python 128
|
|
|
19
19
|
|
|
20
20
|
---
|
|
21
21
|
|
|
22
|
+
## Asymmetric Hybrid Architecture Benchmark (Galaxy S25 - Snapdragon 8 Elite)
|
|
23
|
+
|
|
24
|
+
> **Test Audio**: JFK Inaugural Address (60.59s mono 16kHz WAV)
|
|
25
|
+
> **Silicon**: Qualcomm Snapdragon 8 Elite (Adreno 830 GPU + Oryon CPU)
|
|
26
|
+
|
|
27
|
+
| Model | Pure CPU (4T) | Pure GPU (Vulkan) | Hybrid GPU-CPU (4T) | Hybrid Efficiency (1T) | Speedup / Advantage |
|
|
28
|
+
| :--- | :---: | :---: | :---: | :---: | :--- |
|
|
29
|
+
| **Whisper Tiny** | 4.59s | 9.56s | **4.38s** | 5.48s | **+4.6% vs CPU** (Dispatch overhead minimized) |
|
|
30
|
+
| **Whisper Small** | 15.48s | 16.36s | **12.93s** | 23.02s | **+21.0% vs GPU**, **+16.5% vs CPU** |
|
|
31
|
+
| **Large-v3-Turbo** | 115.05s | 87.87s | **86.52s** | 90.39s | **-28.5s vs CPU**, CPU Load 0% during 80s encode |
|
|
32
|
+
|
|
33
|
+
*Multi-SoC Verification: Tested across Qualcomm Adreno 830, ARM Mali-G78 (Galaxy S21), and ARM Mali-G77 (Galaxy S20).*
|
|
34
|
+
*Upstream Contribution: Official PR submitted to `ggml-org/whisper.cpp` ([PR #4089](https://github.com/ggml-org/whisper.cpp/pull/4089)).*
|
|
35
|
+
|
|
36
|
+
---
|
|
37
|
+
|
|
22
38
|
## Installation & Quickstart
|
|
23
39
|
|
|
24
40
|
### Python (PyPI)
|
|
@@ -28,8 +44,8 @@ pip install termux-stt
|
|
|
28
44
|
```python
|
|
29
45
|
from termux_stt import create_engine
|
|
30
46
|
|
|
31
|
-
# 1. Initialize Engine
|
|
32
|
-
engine = create_engine("whisper", model="
|
|
47
|
+
# 1. Initialize Engine with Hybrid GPU-Encoder / CPU-Decoder Acceleration
|
|
48
|
+
engine = create_engine("whisper", model="small", lang="en", threads=4, split_mode=True)
|
|
33
49
|
|
|
34
50
|
# 2. Transcribe Audio directly into Subtitles
|
|
35
51
|
result = engine.transcribe("samples/jfk_1min.wav")
|
|
@@ -44,6 +60,18 @@ for seg in diar_result.segments:
|
|
|
44
60
|
|
|
45
61
|
```
|
|
46
62
|
|
|
63
|
+
### CLI Command Line Usage
|
|
64
|
+
```bash
|
|
65
|
+
# 1. High-Performance Hybrid Transcribe (Vulkan GPU Encoder + 4 CPU Threads Decoder)
|
|
66
|
+
termux-stt transcribe samples/jfk_1min.wav --model small --split-mode
|
|
67
|
+
|
|
68
|
+
# 2. Ultra-Low Power Background Transcribe (GPU Encoder + 1 CPU Thread Decoder)
|
|
69
|
+
termux-stt transcribe meeting.wav --model small --optimize-1
|
|
70
|
+
|
|
71
|
+
# 3. Force Pure CPU Execution
|
|
72
|
+
termux-stt transcribe samples/jfk_1min.wav --device cpu --threads 4
|
|
73
|
+
```
|
|
74
|
+
|
|
47
75
|
### Node.js / TypeScript (npm)
|
|
48
76
|
```bash
|
|
49
77
|
npm install termux-stt
|
package/README.pypi.md
CHANGED
|
@@ -17,8 +17,8 @@ pip install termux-stt
|
|
|
17
17
|
```python
|
|
18
18
|
from termux_stt import create_engine
|
|
19
19
|
|
|
20
|
-
# 1. Initialize Engine
|
|
21
|
-
engine = create_engine("whisper", model="
|
|
20
|
+
# 1. Initialize Engine with Hybrid GPU-Encoder / CPU-Decoder Acceleration
|
|
21
|
+
engine = create_engine("whisper", model="small", lang="en", threads=4, split_mode=True)
|
|
22
22
|
|
|
23
23
|
# 2. Transcribe Audio directly into Subtitles
|
|
24
24
|
result = engine.transcribe("samples/jfk_1min.wav")
|
|
@@ -36,6 +36,12 @@ for seg in diar_result.segments:
|
|
|
36
36
|
## Description
|
|
37
37
|
Integrates Whisper.cpp, Vosk, and Sherpa-ONNX with a closed-form pure-Python 128-dimensional X-Vector clustering algorithm that operates in under 80MB RAM with zero cloud egress.
|
|
38
38
|
|
|
39
|
+
## Hybrid Performance Highlights (Galaxy S25 - Snapdragon 8 Elite)
|
|
40
|
+
|
|
41
|
+
- **Whisper Small (60s JFK Audio)**: **12.93s** wall-clock time (**21.0% faster than Pure GPU**, **16.5% faster than Pure CPU**).
|
|
42
|
+
- **Whisper Large-v3-Turbo**: **86.52s** wall-clock time (28.5s faster than Pure CPU), with zero host CPU load during the 80s audio encoding stage.
|
|
43
|
+
- Upstream PR submitted to `ggml-org/whisper.cpp` ([PR #4089](https://github.com/ggml-org/whisper.cpp/pull/4089)).
|
|
44
|
+
|
|
39
45
|
## Documentation
|
|
40
46
|
- [Official Documentation & API Reference](https://uno-km.vercel.app/lib/stt/)
|
|
41
47
|
- [GitHub Repository](https://github.com/uno-km/termux-stt)
|
package/package.json
CHANGED