README_en.md

Android

简体中文 | English

Disclaimer, this document was obtained through machine translation, please check the original document here.

Convert model

1. To convert your models, run convert-ggml.py from the root of your Whisper-Finetune project to convert your models to ggml format for your Android project. The models you need to convert can be original Transformers. It can also be a fine-tuned model.

python convert-ggml.py --model_dir=models/whisper-tiny-finetune/ --output_path=models/ggml-model.bin

2. Put the model in the Android project app/SRC/main/assets/models directory, and then you can use the Android open Studio project.

Build notes

1. The default NDK version used is 25.2.9519653, if you change the other version below, you will need to change the configuration in app/build.gradle.
2. Note that in real use, be sure to release the release APK package so that inference is fast.
3. This project has released the release APK package, please scan the code at the end of the Whisker-finetune project homepage to download it.

Effect picture

The effect picture is as follows. The model used here is quantized as a half-precision tiny model, which has a low accuracy.

Download APK

Can click here to download the Android APK, note that in order to install package is small, quantitative model used here for half a tiny model precision and accuracy is not high, if you want to change model, please compile the project execution.