Show HN: CPU-only transcription for YouTube, TikTok, X, Instagram videos

hckrnws

Show HN: CPU-only transcription for YouTube, TikTok, X, Instagram videos

by mrkn1

piotrrojek

32d

If someone is interested, this is my supershort zsh/bash scripts that I keep in .zshrc for doing the same thing using plain whisper.cpp, ffmpeg and yt-dlp (`brew install whisper-cpp yt-dlp` for Mac); I output it in vtt format (subtitles) though, but it's easy enough to change it to txt.

  yt_to_srt() {
    local url="$1"
    local output_base="$2"
    local language="${3:-en}"

    yt-dlp -x --audio-format wav --postprocessor-args "-ar 16000" -o "$output_base.wav" "$url"
    whisper-cli --language "$language" --model "$WHISPER_MODEL" --split-on-word --max-len 65 --output-vtt --output-file "$output_base" --file "$output_base.wav"
    rm "$output_base.wav"
  }

  file_to_srt() {
    local filepath="$1"
    local language="${2:-en}"

    local filename=$(basename "$filepath")
    local filename_no_ext="${filename%.*}"
    local output_base="$filename_no_ext"
    local temp_wav="$output_base.wav"

    ffmpeg -i "$filepath" -vn -acodec pcm_s16le -ar 16000 -ac 1 "$temp_wav"
    whisper-cli --language "$language" --model "$WHISPER_MODEL" --split-on-word --max-len 65 --output-vtt --output-file "$output_base" --file "$temp_wav"
    rm "$temp_wav"
  }

plus additional bootstrap script for large-v3-turbo model from my chez-moi dotfiles:

  #!/bin/bash
  # Download whisper.cpp models from Hugging Face (runs once per machine).
  set -euo pipefail
  MODELS_DIR="$HOME/whisper-models"
  BASE_URL="https://huggingface.co/ggerganov/whisper.cpp/resolve/main"
  MODELS=("ggml-large-v3-turbo.bin" "ggml-tiny.bin")
  mkdir -p "$MODELS_DIR"
  for model in "${MODELS[@]}"; do
    if [ ! -f "$MODELS_DIR/$model" ]; then
      echo "Downloading $model..."
      curl -L --progress-bar -o "$MODELS_DIR/$model" "$BASE_URL/$model"
    else
      echo "$model already exists, skipping."
    fi
  done
  echo "Whisper models ready at $MODELS_DIR"

ramon156
12d
yt-dlp can download auto-subtitles and regular subtitles, why not do that and fall back to whisper?
1. piotrrojek
  02d
  To be frank I didn't know there's such an option :-)

delis-thumbs-7e
12d
Am I a bit thick, but first we created this amazing way to transfer any text very cheaply and fast over network, then we (well, I think it was Meta and Google) decided that no, everything must be a video, then we added subtitles and AI-transcriptions to those videos and now we just dowload transcriptions of those videos presumably to feel LLM to make summaries of them in order to… Read. Them.
I think I’m gonna go read a book.
1. mrkn1 • OP
  02d
  Good point! I haven't found a faster way to consume info than reading. But depends on the type of learner you are (visual, auditory, hands-on/interactive, etc)
spudlyo
83d
So, this project consists of a ~175 line README and a ~500 line Python program that glues yt-dlp and Kroko together. Neat.
I guess if it encourages you to install and figure out how to use ffmpeg, yt-dlp, kroko, numpy, and onnx that's a good thing. Sometimes just knowing a thing is possible is a huge benefit.
1. iririririr
  13d
  I see the value as a centralized anti-content-blocker.
  This repo is now a good way to centralize hacks around the sure-to-come blockers those platforms will add to prevent download.
  Just like uBlockOrigin was a way to centralize all the "just run this greasemonkey script" comments, I can see this getting a huge following for people who really value transcriptions.
throw98226
12d
Works extremely well. Command to install on Debian 13:
sudo apt update && sudo apt install -y ffmpeg python3-pip python3-venv && git clone https://github.com/kouhxp/yapsnap.git && cd yapsnap && python3 -m venv ~/yapsnap-venv && source ~/yapsnap-venv/bin/activate && pip install --upgrade pip && pip install .
On a 32GB ThinkPad X13, a 21 minute YouTube video was processed by yapsnap under 2 minutes.
Very well done!
1. mrkn1 • OP
  02d
jorritpr
12d
Very cool, I'm also working on a captioning/subtitling project for the lecture recordings for the university I work at.
My biggest challenge is finding a proper language model that is fast enough and accurate enough since I have to caption about 600 hours of video per week and I preferably want to run all of this on a tiny server (2 cores 4 GB memory). This tool could easily do that with the kroko model but I'll have to test if the accuracy is good enough.
Also in my own scripts I'm using ffmpeg to download just the audio of the videos that I want to caption, which saves a lot of bandwith and speeds up the whole process. As far as I can see this tool doesn't do that, that would be a nice functionality to add, plus an option to turn the output into a working .srt file.
1. mrkn1 • OP
  02d
7777777phil
22d
Tis is very simple and very cool! Just installed it on my Hetzner box where I run a remote controlled local agent so now I can basically chat/email a video link to get a summary and/or ask questions. The only issue was YouTube's PO Token requirement (web/mweb clients refuse to serve formats from datacenter IPs without a valid Proof-of-Origin token.) So I had to find a client that still work without PO Token first. Thanks for sharing!
1. mrkn1 • OP
  12d
  thank you! good use case, what hetzner box specs have you chosen?
niraj-agarwal
13d
Had Claude test it out on 3 videos. Worked at 5-8x realtime. The beauty of it is that it works on all videos, not just the one with transcripts. Combine it with YouTube search and LLM takeaways from transcripts, and you have super-efficient content consumption. There are SaaS products that charge 1 cent per video for those with transcripts. There is a viable product in here somewhere, methinks.
1. mrkn1 • OP
  03d
  thanks for running it Niraj. I see something similar on my machine, which still surprises me every time lol
HDBaseT
22d
Wouldn't it still be more efficient to do GPU transcriptions anyways? is this something we could actually put the effectively useless NPUs to use in modern laptops?
1. dharma1
  02d
  yes GPU is significantly faster, but cpu only lets you do it anywhere - wasm in the browser, any server etc.
  NPUs - definitely a good use case for at least part of it, there are ports of whisper that use coreML/ANE with less power and 3x speed of CPU only
majorchord
12d
I thought ONNX models were only for text-to-speech? How does one tell them apart if I find some files online?
1. mrkn1 • OP
  02d
  [dead]
canadiantim
22d
Nice. Can it do speaker diarization?
1. mrkn1 • OP
  02d
  will work on it, that would be neat. I love pyannote but not happening on CPU at reasonable speeds lol
ranger_danger
12d
How is this so much faster than even GPU-based whisper?
1. mrkn1 • OP
  02d
  small, ONNX-optimized models designed specifically for low-latency CPU streaming, so it avoids overhead of large transformer arch and GPU memory transfers
charcircuit
33d
Most of these platforms already have transcriptions built in.
1. mrkn1 • OP
  23d
  Youtube has transcripts on most videos, not all. The others don't expose them. If you mean the "transcript APIs" for TikTok/IG/X, they are all transcribing audio like yapsnap does. If you have a way to pull native ones, let me know, genuinely curious.
ranger_danger
23d
How can we transcribe other languages besides English?
1. mrkn1 • OP
  13d
  Just download the model for your preferred language, all hosted on the Kroko-ASR collection here: https://huggingface.co/Banafo/Kroko-ASR/tree/main Right now you have Dutch, French, Portuguese, Spanish, German, Italian, Swedish, Swiss German, Hebrew, and Turkish. Grab the one that matches your audio, point yapsnap at it with --model (or set KROKO_MODEL), and you're set!

mrkn1 • OP

02d

Added Diarization / Speaker Separation that is fast and CPU only. Thank you all for the great feedback and support. PRs welcome!

  yapsnap "https://www.youtube.com/watch?v=NzKJ-xO-VhE" --diarize

  SPEAKER_00 [00:00]: Welcome to the show.
  SPEAKER_01 [00:03]: Glad to be here, thanks for having me.
  SPEAKER_00 [00:08]: Let's get started.

dubaiamir1111
01d
dubaiamir1111
dmos62
32d
Now make it distinguish speakers and we really have something. As far as I know, that's significantly harder though.
1. mrkn1 • OP
  12d
  done! new version separates speakers on CPU fast
alextmx
01d
[dead]
photonair
02d
[dead]
xnx
03d
[dead]
chris_explicare
02d
[dead]

yt_to_srt() { local url="$1" local output_base="$2" local language="${3:-en}" yt-dlp -x --audio-format wav --postprocessor-args "-ar 16000" -o "$output_base.wav" "$url" whisper-cli --language "$language" --model "$WHISPER_MODEL" --split-on-word --max-len 65 --output-vtt --output-file "$output_base" --file "$output_base.wav" rm "$output_base.wav" } file_to_srt() { local filepath="$1" local language="${2:-en}" local filename=$(basename "$filepath") local filename_no_ext="${filename%.*}" local output_base="$filename_no_ext" local temp_wav="$output_base.wav" ffmpeg -i "$filepath" -vn -acodec pcm_s16le -ar 16000 -ac 1 "$temp_wav" whisper-cli --language "$language" --model "$WHISPER_MODEL" --split-on-word --max-len 65 --output-vtt --output-file "$output_base" --file "$temp_wav" rm "$temp_wav" }

#!/bin/bash # Download whisper.cpp models from Hugging Face (runs once per machine). set -euo pipefail MODELS_DIR="$HOME/whisper-models" BASE_URL="https://huggingface.co/ggerganov/whisper.cpp/resolve/main" MODELS=("ggml-large-v3-turbo.bin" "ggml-tiny.bin") mkdir -p "$MODELS_DIR" for model in "${MODELS[@]}"; do if [ ! -f "$MODELS_DIR/$model" ]; then echo "Downloading $model..." curl -L --progress-bar -o "$MODELS_DIR/$model" "$BASE_URL/$model" else echo "$model already exists, skipping." fi done echo "Whisper models ready at $MODELS_DIR"

yapsnap "https://www.youtube.com/watch?v=NzKJ-xO-VhE" --diarize SPEAKER_00 [00:00]: Welcome to the show. SPEAKER_01 [00:03]: Glad to be here, thanks for having me. SPEAKER_00 [00:08]: Let's get started.