Transcribe Audio to Text Locally

git clone https://github.com/ggerganov/whisper.cpp
cd whisper.cpp
bash ./models/download-ggml-model.sh large-v3


python -m venv whisper-env
source whisper-env/bin/activate
pip3 install ane_transformers openai-whisper coremltools
./models/generate-coreml-model.sh large-v3
brew install cmake
WHISPER_COREML=1 make -j

nano ~/.zshrc
source ~/.zshrc

function transcribe_ru() {
    if [ -z "$1" ]; then
        echo "Usage: transcribe_ru <audio_file> [model]"
        echo "Default model: large-v3-turbo"
        return 1
    fi

    local input_file="$1"
    local filename=$(basename "$input_file")
    local stem="${filename%.*}"
    local ext="${filename##*.}"
    local model="${2:-large-v3}" 
    local whisper_path="$HOME/Documents/projects/ai-whisper/whisper.cpp"
    local whisper_bin="$whisper_path/build/bin/whisper-cli"
  
    # Check if model exists
    if [ ! -f "$whisper_path/models/ggml-${model}.bin" ]; then
        echo "Error: Model 'ggml-${model}.bin' not found in $whisper_path/models/"
        return 1
    fi

    # Only convert if not WAV
    local audio_file="$input_file"

    if [ "$ext" != "wav" ] && [ "$ext" != "WAV" ]; then
        echo "Converting ${ext} to 16kHz mono WAV..."
        audio_file="/tmp/${stem}_16k.wav"
        ffmpeg -y -i "$input_file" -ar 16000 -ac 1 -c:a pcm_s16le "$audio_file" > /dev/null 2>&1
        if [ ! -f "$audio_file" ]; then
            echo "Error: Audio conversion failed."
            return 1
        fi
    else
        echo "Using existing WAV file (assuming 16kHz mono)..."
    fi

    echo "Transcribing (Language: Russian, Model: $model, VAD: ON)..."
    "$whisper_bin" \
        -m "$whisper_path/models/ggml-${model}.bin" \
        -f "$audio_file" \
        -l ru \
        -otxt \
        -of "${stem}"
    echo "✅ Done! Output: ${stem}.txt"
    
    # Cleanup temp file if created
    if [ "$audio_file" != "$input_file" ]; then
        rm "$audio_file"
    fi
}

ffmpeg -i input.mkv -ar 16000 -ac 1 -c:a pcm_s16le output.wav