Generate summaries of audio data from audio files, urls, and youtube videos.
Two command-line scripts are provided: get-summary.py and get-text.py.
The Whisper model is used for speech-to-text and GPT-3.5/GPT-4 is used for
generating summaries.
get-summary.py can be used to generate summaries for any text.
For example, python src/get-summary.py -i path/to/audio.mp3 will print out
a summary of the speech
First, clone this repo
git clone --recurse-submodules https://github.com/Dando18/audio-summarizer.gitThen build whisper.cpp.
cd tpl/whisper.cpp
# build whisper
# cublas, opencl, coreml, etc.
# see https://github.com/ggerganov/whisper.cpp for backends
WHISPER_OPENBLAS=1 make -j
# download model(s)
# current options: tiny.en tiny base.en base small.en small medium.en medium
# large-v1 large-v2 large-v3
# see https://github.com/ggerganov/whisper.cpp for all options
make large-v3To get a summary for my-audio.wav you can run:
# create a cache for saving intermediate data
mkdir -p cache
python src/get-summary.py \
--input my-audio.wav \
--output summary.txt \
--model-path tpl/whisper.cpp/models/ggml-large-v3.bin \
--n-threads 4 \
--prompt 'Summarize the following text: {text}' \
--max-tokens 4096 \
--cache-dir cache \
--openai-model gpt-4-0125-preview
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent