rmads is a CLI for removing ads from audio files to quantify ad statistics
sudo dnf install python3-pip mp3splt ffmpeg OR sudo apt install python3-pip mp3splt ffmpeg
pip install -U -r requirements.txt
To use gemini, you must have a Gemini API key
defined as GEMINI_API_KEY="YOUR_API_KEY" in an .env file
pytest -v
usage: rmads.py [-h]
[-a {Meta-Llama-3-8B-Instruct.Q4_0.gguf,Nous-Hermes-2-Mistral-7B-DPO.Q4_0.gguf,Phi-3-mini-4k-instruct.Q4_0.gguf,orca-mini-3b-gguf2-q4_0.gguf,gpt4all-13b-snoozy-q4_0.gguf}]
[-c] [-d DIRECTORY] [-e THRESHOLD]
[-g {gemini-pro,gemini-1.0-pro,gemini-1.5-pro,gemini-1.5-flash,gemini-1.5-flash-8b,gemini-2.0-flash-exp}]
[-G {gemini-1.5-pro,gemini-1.5-flash,gemini-1.5-flash-8b,gemini-2.0-flash-exp}] [-k keywords.txt]
[-l LANGUAGE] [-m SECONDS] [-p] [-P] [-r [SEGMENT ...]] [--rpm RPM] [-s SHOTS] [-t [SEGMENT ...]]
[-w {tiny,tiny.en,base,base.en,small,small.en,medium,medium.en,large}] [-v]
audiofile [audiofile ...]
rmads is a CLI for removing ads from audio files to quantify ad statistics
positional arguments:
audiofile audio file to remove ads from
options:
-h, --help show this help message and exit
-a {Meta-Llama-3-8B-Instruct.Q4_0.gguf,Nous-Hermes-2-Mistral-7B-DPO.Q4_0.gguf,Phi-3-mini-4k-instruct.Q4_0.gguf,orca-mini-3b-gguf2-q4_0.gguf,gpt4all-13b-snoozy-q4_0.gguf}, --gpt4all {Meta-Llama-3-8B-Instruct.Q4_0.gguf,Nous-Hermes-2-Mistral-7B-DPO.Q4_0.gguf,Phi-3-mini-4k-instruct.Q4_0.gguf,orca-mini-3b-gguf2-q4_0.gguf,gpt4all-13b-snoozy-q4_0.gguf}
gpt4all model to use for ad recognition (default: Meta-Llama-3-8B-Instruct.Q4_0.gguf)
-c, --count count the number of split files created and then exit (default: False)
-d DIRECTORY, --dir DIRECTORY
working directory (default: .)
-e THRESHOLD, --th THRESHOLD
dB threshold level (-96 to 0) for silence when splitting audio (default: -48)
-g {gemini-pro,gemini-1.0-pro,gemini-1.5-pro,gemini-1.5-flash,gemini-1.5-flash-8b,gemini-2.0-flash-exp}, --gemini {gemini-pro,gemini-1.0-pro,gemini-1.5-pro,gemini-1.5-flash,gemini-1.5-flash-8b,gemini-2.0-flash-exp}
gemini model to use for ad recognition (default: None)
-G {gemini-1.5-pro,gemini-1.5-flash,gemini-1.5-flash-8b,gemini-2.0-flash-exp}, --gemini-audio {gemini-1.5-pro,gemini-1.5-flash,gemini-1.5-flash-8b,gemini-2.0-flash-exp}
gemini model to use for audio upload ad recognition (default: None)
-k keywords.txt, --keyword-file keywords.txt
line separated keyword file to use to id an ad (default: None)
-l LANGUAGE, --lang LANGUAGE
language to use for audio to text (default: en)
-m SECONDS, --min SECONDS
minimum seconds (> 0.0) to be considered valid silence when splitting audio (default: 1.0)
-p, --purge purge all progress files of file arg (default: False)
-P, --purge-all purge all progress files (default: False)
-r [SEGMENT ...], --retry [SEGMENT ...]
split segment to retry (01, 02, ...) (default: None)
--rpm RPM override requests per minute when making API calls (default: None)
-s SHOTS, --shots SHOTS
shots (> 0) of non silence when splitting audio (default: 25)
-t [SEGMENT ...], --toggle [SEGMENT ...]
split segment to toggle ad (01, 02, ...) (default: None)
-w {tiny,tiny.en,base,base.en,small,small.en,medium,medium.en,large}, --whisper {tiny,tiny.en,base,base.en,small,small.en,medium,medium.en,large}
whisper model to use for text recognition (default: base.en)
-v, --verbose verbose output (default: None)
Change -e, -m or -s to adjust number of split files. Change -w to adjust audio to text recognition. Change -a or -g to
adjust ad recognition.
./src/rmads.py -d tmp tests/road_not_taken.mp3
audio="tests/road_not_taken.mp3" min=1.0 shots=25 th=-48 splits=5 whisper="base.en" llm="Meta-Llama-3-8B-Instruct.Q4_0.gguf"
==========
Generating text from "road_not_taken_silence_01.mp3"...
Calling gpt4all using "road_not_taken_silence_01.txt"...
Response = YES
==========
Generating text from "road_not_taken_silence_02.mp3"...
Calling gpt4all using "road_not_taken_silence_02.txt"...
Response = NO
==========
Generating text from "road_not_taken_silence_03.mp3"...
Calling gpt4all using "road_not_taken_silence_03.txt"...
Response = YES
==========
Generating text from "road_not_taken_silence_04.mp3"...
Calling gpt4all using "road_not_taken_silence_04.txt"...
Response = NO
==========
Generating text from "road_not_taken_silence_05.mp3"...
Calling gpt4all using "road_not_taken_silence_05.txt"...
Response = NO
==========
Total ads = 2
Total ad time = 0:00:16 of 0:00:59 (27.5%)
Ads per minute = 2.03
Average ads = 1 per 0:00:29
./src/rmads.py -d tmp tests/road_not_taken.mp3 -p -g gemini-pro
Purged "tests/road_not_taken.mp3" progress files in "tmp"
audio="tests/road_not_taken.mp3" min=1.0 shots=25 th=-48 splits=5 whisper="base.en" llm="gemini-pro"
==========
Generating text from "road_not_taken_silence_01.mp3"...
Calling gemini using "road_not_taken_silence_01.txt"...
Response = YES
==========
Generating text from "road_not_taken_silence_02.mp3"...
Calling gemini using "road_not_taken_silence_02.txt"...
Response = NO
==========
Generating text from "road_not_taken_silence_03.mp3"...
Waiting for 0.3 seconds to call gemini-pro because rpm = 15
Calling gemini using "road_not_taken_silence_03.txt"...
Response = YES
==========
Generating text from "road_not_taken_silence_04.mp3"...
Waiting for 0.3 seconds to call gemini-pro because rpm = 15
Calling gemini using "road_not_taken_silence_04.txt"...
Response = NO
==========
Generating text from "road_not_taken_silence_05.mp3"...
Calling gemini using "road_not_taken_silence_05.txt"...
Response = NO
==========
Total ads = 2
Total ad time = 0:00:16 of 0:00:59 (27.5%)
Ads per minute = 2.03
Average ads = 1 per 0:00:29
gemini audio recognition1
./src/rmads.py -d tmp tests/road_not_taken.mp3 -G gemini-2.0-flash-exp
Calling gemini using "tests/road_not_taken.mp3"...
Response =
00:00:00.000 00:00:03.000 - not ad
00:00:03.000 00:00:22.000 - not ad
00:00:25.000 00:00:26.000 - not ad
00:00:26.000 00:00:48.000 - not ad
==========
Total ads = 1
Total ad time = 0:00:14 of 0:00:59 (24.4%)
Ads per minute = 1.02
Average ads = 1 per 0:00:59
Footnotes
-
Using gemini for audio upload ad recognition is not quite accurate yet ↩