Whispy

#!/bin/bash
# Combined Whisper transcription script
# Usage:
#   whispy.sh [--cpu|--gpu] <audio-file> [model]     - Transcribe a single file
#   whispy.sh [--cpu|--gpu] --all [model]             - Batch transcribe all MP3 files in current directory

MODEL_DEFAULT="medium"
USE_CPU=true

# Parse CPU/GPU flag
if [ "$1" = "--cpu" ]; then
    USE_CPU=true
    shift
elif [ "$1" = "--gpu" ]; then
    USE_CPU=false
    shift
fi

# Set CUDA visibility based on flag
if [ "$USE_CPU" = true ]; then
    export CUDA_VISIBLE_DEVICES=""
    echo "Using CPU mode"
else
    echo "Using GPU mode"
fi

# Function to transcribe a single file
transcribe_file() {
    local FILE="$1"
    local MODEL="$2"
    local BASE="${FILE%.mp3}"
    local TXT="$BASE.txt"
    local SRT="$BASE.srt"

    if [ ! -f "$FILE" ]; then
        echo "Error: File '$FILE' not found"
        return 1
    fi

    # Skip if both transcription files already exist
    if [ -f "$TXT" ] && [ -f "$SRT" ]; then
        echo "Skipping '$FILE' - already transcribed ($TXT and $SRT exist)"
        return 0
    fi

    echo "Transcribing '$FILE' with model '$MODEL'..."

    source "$HOME/.venvs/whisper-venv/bin/activate"

    python3 - "$FILE" "$MODEL" "$TXT" "$SRT" <<'PY'
import sys, os
import whisper
from datetime import timedelta

audio_file, model_name, txt_path, srt_path = sys.argv[1:]

model = whisper.load_model(model_name)
result = model.transcribe(audio_file, verbose=True, language=None)

print(f"\nDetected language: {result['language']}")

# Write plain text
with open(txt_path, "w", encoding="utf-8") as txt:
    txt.write(result["text"])

# Write SRT
with open(srt_path, "w", encoding="utf-8") as srt:
    for i, segment in enumerate(result["segments"], 1):
        start = str(timedelta(seconds=int(segment['start']))).split('.')[0]
        end = str(timedelta(seconds=int(segment['end']))).split('.')[0]
        srt.write(f"{i}\n")
        srt.write(f"{start} --> {end}\n")
        srt.write(f"{segment['text'].strip()}\n\n")

print(f"\nDone → {os.path.basename(txt_path)} + {os.path.basename(srt_path)}")
PY

    if [ $? -eq 0 ]; then
        echo -e "\033[0;32mFinished:\033[0m $TXT  $SRT"
        return 0
    else
        return 1
    fi
}

# Function to batch process all MP3 files
batch_transcribe() {
    local MODEL="$1"
    local PROCESSED=0
    local FAILED=0

    echo "Starting batch transcription with model: $MODEL"
    echo "================================================"
    echo ""

    # Loop through all MP3 files
    for mp3_file in *.mp3; do
        # Skip if no MP3 files found
        if [ "$mp3_file" = "*.mp3" ]; then
            echo "No MP3 files found in current directory"
            exit 1
        fi

        # Check if already transcribed (skip if .txt exists)
        base_name="${mp3_file%.mp3}"
        if [ -f "$base_name.txt" ]; then
            echo "⏭️  Skipping: $mp3_file (already transcribed)"
            continue
        fi

        echo "🎙️  Processing: $mp3_file"
        echo "---"

        if transcribe_file "$mp3_file" "$MODEL"; then
            ((PROCESSED++))
            echo "✅ Success: $base_name.txt + $base_name.srt"
        else
            ((FAILED++))
            echo "❌ Failed: $mp3_file"
        fi

        echo ""
    done

    echo "================================================"
    echo "Batch transcription complete!"
    echo "  Processed: $PROCESSED"
    echo "  Failed: $FAILED"
    echo "  Total: $((PROCESSED + FAILED))"
}

# Main script logic
if [ "$1" = "--all" ]; then
    # Batch mode
    MODEL="${2:-$MODEL_DEFAULT}"
    batch_transcribe "$MODEL"
elif [ -z "$1" ]; then
    # No arguments - show usage
    echo "Usage:"
    echo "  whispy.sh [--cpu|--gpu] <audio-file> [model]    - Transcribe a single file"
    echo "  whispy.sh [--cpu|--gpu] --all [model]            - Batch transcribe all MP3 files"
    echo ""
    echo "Options:"
    echo "  --cpu    Use CPU-only mode (default)"
    echo "  --gpu    Use GPU acceleration"
    echo ""
    echo "Models: tiny, base, small, medium (default), large"
    exit 1
else
    # Single file mode
    FILE="$1"
    MODEL="${2:-$MODEL_DEFAULT}"
    transcribe_file "$FILE" "$MODEL"
fi