My Script for STT

comments.md

Script Purpose:

This script transcribes audio into text using either a remote Groq API or a local Whisper.cpp model based on user input. Triggered via keybindings, it supports multiple languages.

Keybindings:

• whpr → Recognize audio in English.
• whpr fr → Recognize audio in French.
• whpr es → Recognize audio in Spanish.
• whpr lang → Recognize audio in any supported language.

Workflow:

1. Recording:

   • Starts recording audio using rec and saves the file to /dev/shm for faster RAM-based processing.
   • Stops recording when triggered again.

2. Processing:

   • If using Groq API (default), it sends the audio for transcription. Falls back once if the request fails.
   • If using Whisper.cpp locally, it processes the audio using the specified model.

3. Output:

   • The transcribed text is passed to the keyboard using xdotool.
   • Notifications (notify-send) provide status updates.

Configuration:

• Groq API KEY: Place it at ~/groq.token.txt.
• Local Setup:
  • Set REMOTE=false to use Whisper.cpp.
  • Install Whisper.cpp and download the models:
    • Multilingual: /home/user/tmp/whisper.cpp/models/ggml-small.bin
    • English-only: /home/user/tmp/whisper.cpp/models/ggml-small.en.bin
  • Adjust paths as needed.

Dependencies:

• rec, xdotool, notify-send (common in Ubuntu/Debian-based systems).
• Groq API or Whisper.cpp for transcription.

Notes:

• Tested on Xubuntu; It should work almost out-of-the-box with ubuntu based and debian based distros, compatibility may vary on other distributions.
• Ensure required tools are installed for xdotool (keyboard input) and notify-send (notifications).

whpr

#!/bin/bash
ramf="/dev/shm/sfile.wav"
language="en"
# this is not taken into account for X11
GROQ_API_KEY=`cat ~/groq.token.txt`
REMOTE=true

# Check if a language parameter is provided
if [ ! -z "$1" ]; then
    language="$1"
fi

# Check if the process is running
if ps aux | grep -v grep | grep "$ramf" > /dev/null; then
    time=$(ffprobe -v error -show_entries format=duration -of default=noprint_wrappers=1:nokey=1 "$ramf" | awk '{printf "%.1f\n", $1}')
    notify-send "Whisper process" "Processing $time seconds"
    PID=$(ps aux | grep "$ramf" | grep -v grep | awk '{print $2}')
    kill $PID
    start_time=$(date +%s)
    if [ "$REMOTE" = true ]; then
        sleep 0.05
        str=$(curl https://api.groq.com/openai/v1/audio/transcriptions \
                   -H "Authorization: Bearer $GROQ_API_KEY" \
                   -H "Content-Type: multipart/form-data" \
                   -F file="@$ramf" \
                   -F language="$language" \
                   -F model="whisper-large-v3" | jq -r .text)
        if [ "$str" = "null" ] || [ -z "$str" ]; then
            str=$(curl https://api.groq.com/openai/v1/audio/transcriptions \
                       -H "Authorization: Bearer $GROQ_API_KEY" \
                       -H "Content-Type: multipart/form-data" \
                       -F file="@$ramf" \
                       -F language="$language" \
                       -F model="whisper-large-v3" | jq -r .text)
        fi
    else
        if [ "$language" != "en" ]; then
            str="$(/home/user/tmp/whisper.cpp/build/bin/whisper-cli -m /home/user/tmp/whisper.cpp/models/ggml-small.bin -l "$language" -nt -fa -t 6 -f $ramf 2>/dev/null)"
        else
            str="$(/home/user/tmp/whisper.cpp/build/bin/whisper-cli -m /home/user/tmp/whisper.cpp/models/ggml-small.en.bin -nt -fa -t 6 -f $ramf 2>/dev/null)"
        fi
        notify-send "Processed in $lapse_time seconds" ""
    fi
    end_time=$(date +%s)
    lapse_time=$((end_time - start_time))
    str2=$(echo "$str" | tr -d '\n' | sed 's/^[ \t]*//;s/[ \t]*$//')
    echo -n "$str2" | xdotool type --clearmodifiers --delay 0 --file -
    rm "$ramf"
else
    notify-send "Starting Record ($language)" ""
    # Start recording
    nohup rec -q -t wav $ramf rate 16k channels 1 2>/dev/null &
fi