FreePBX Transcription using Whisper

FreePBX Transcription Using Whisper AI (Offline and Secure)

We’ve integrated automated transcription into our FreePBX systems using OpenAI’s Whisper engine. Everything runs locally with no third-party APIs and no data ever leaves your server.

Our setup includes:

• setup_transcriber.sh: Installs Whisper in a clean, self-contained environment without affecting FreePBX or system packages

• transcribe_watcher.sh: Monitors call recordings and automatically transcribes new files for delivery or storage

This solution gives you:

• Instant transcription of voicemails or call recordings

• Full privacy with no cloud services

• Searchable text logs for audits, customer service, or documentation

There are no licensing fees or API limits. Just simple, reliable transcription on your own system.

This is now available as part of our managed VoIP service. If you need this set up or want to learn more, contact us.

The setup installs the transcribe dependencies and all is stored in /var/transcripts. It runs independently from FreePBX. The watcher service monitors the recordings and transcribes the audio then sends it to the email on the extension and also an archive@email.com which is manually set. You do need an smtp to send emails.

Installer (setup_transcriber.sh)

#!/bin/bash

set -euo pipefail

# === CONFIGURATION ===

PYTHON_VERSION="3.10.14"

OPENSSL_VERSION="1.1.1l" # Correctly defined here

INSTALL_PREFIX="/opt"

OPENSSL_PREFIX="$INSTALL_PREFIX/openssl-$OPENSSL_VERSION"

PYTHON_PREFIX="$INSTALL_PREFIX/python-$PYTHON_VERSION"

SRC_DIR="/usr/local/src"

WHISPER_ENV="/opt/whisper_env"

LOG_FILE="/var/log/setupdiagnostic.log" # New log file

# === LOGGING SETUP ===

# Redirect all stdout and stderr to the log file

exec > >(tee -a "$LOG_FILE") 2>&1

echo "--- Starting Whisper Installation and Service Setup ---"

echo "Log file: $LOG_FILE"

echo "Started at: $(date)"

# Function to check if Python is functional (can import ssl)

check_python_functional() {

if [ -x "$PYTHON_PREFIX/bin/python3.10" ]; then

# Test if the Python binary can import ssl and certifi

if "$PYTHON_PREFIX/bin/python3.10" -c 'import ssl; import certifi; print("Python functional.");' &>/dev/null; then

return 0 # Functional

fi

fi

return 1 # Not functional or not found

}

# === PREREQUISITES AND DEPENDENCY INSTALLATION ===

echo "Checking for Python $PYTHON_VERSION installation..."

if check_python_functional; then

echo "Python $PYTHON_VERSION already installed and functional at $PYTHON_PREFIX. Skipping build."

else

echo "Python $PYTHON_VERSION not found or not functional. Proceeding with installation."

# Install development tools and libraries required for building Python and OpenSSL

echo "Installing system dependencies..."

yum groupinstall -y "Development Tools" || { echo "Error: Failed to install Development Tools. Check $LOG_FILE for details."; exit 1; }

yum install -y gcc zlib-devel wget make openssl-devel libffi-devel bzip2-devel xz-devel || { echo "Error: Failed to install system dependencies. Check $LOG_FILE for details."; exit 1; }

fi

mkdir -p "$SRC_DIR"

cd "$SRC_DIR" || { echo "Error: Failed to change to source directory $SRC_DIR. Check $LOG_FILE for details."; exit 1; }

# === Check and Install inotify-tools ===

echo "Checking for inotify-tools..."

if ! rpm -q inotify-tools &>/dev/null; then

echo "inotify-tools not found. Installing..."

yum install -y inotify-tools || { echo "Error: Failed to install inotify-tools. Check $LOG_FILE for details."; exit 1; }

echo "inotify-tools installed successfully."

else

echo "inotify-tools already installed. Skipping installation."

fi

# === OPENSSL INSTALLATION (Custom Build) ===

echo "Checking for OpenSSL $OPENSSL_VERSION installation..."

if [ ! -d "$OPENSSL_PREFIX" ]; then

echo "OpenSSL $OPENSSL_VERSION not found. Downloading and building."

# Clean up source directory for OpenSSL before extracting

rm -rf "openssl-$OPENSSL_VERSION"

wget -q "https://www.openssl.org/source/openssl-$OPENSSL_VERSION.tar.gz" || { echo "Error: Failed to download OpenSSL source. Check $LOG_FILE for details."; exit 1; }

tar -xf "openssl-$OPENSSL_VERSION.tar.gz" || { echo "Error: Failed to extract OpenSSL source. Check $LOG_FILE for details."; exit 1; }

# FIX: Corrected variable name from OPENSL_VERSION to OPENSSL_VERSION

cd "openssl-$OPENSSL_VERSION" || { echo "Error: Failed to change to OpenSSL source directory. Check $LOG_FILE for details."; exit 1; }

./config --prefix="$OPENSSL_PREFIX" --openssldir="$OPENSSL_PREFIX" shared zlib || { echo "Error: OpenSSL config failed. Check $LOG_FILE for details."; exit 1; }

make -j"$(nproc)" || { echo "Error: OpenSSL make failed. Check $LOG_FILE for details."; exit 1; }

make install || { echo "Error: OpenSSL make install failed. Check $LOG_FILE for details."; exit 1; }

cd ..

echo "OpenSSL $OPENSSL_VERSION installed to $OPENSSL_PREFIX."

else

echo "OpenSSL $OPENSSL_VERSION already found at $OPENSSL_PREFIX."

fi

# === SET ENVIRONMENT FOR BUILDING PYTHON ===

# These exports are crucial for Python to find and link against the custom OpenSSL build

export LD_LIBRARY_PATH="$OPENSSL_PREFIX/lib"

export CPPFLAGS="-I$OPENSSL_PREFIX/include"

export LDFLAGS="-L$OPENSSL_PREFIX/lib"

echo "Environment variables set for Python build (LD_LIBRARY_PATH, CPPFLAGS, LDFLAGS)."

# === PYTHON INSTALLATION (Custom Build) ===

echo "Checking for Python $PYTHON_VERSION installation at $PYTHON_PREFIX..."

if check_python_functional; then

echo "Python $PYTHON_VERSION already installed and functional at $PYTHON_PREFIX. Skipping build."

else

echo "Python $PYTHON_VERSION not found or not functional. Building Python."

echo "Checking for Python $PYTHON_VERSION source archive..."

if [ ! -f "Python-$PYTHON_VERSION.tgz" ]; then

echo "Python $PYTHON_VERSION source not found. Downloading."

wget -q "https://www.python.org/ftp/python/$PYTHON_VERSION/Python-$PYTHON_VERSION.tgz" || { echo "Error: Failed to download Python source. Check $LOG_FILE for details."; exit 1; }

fi

echo "Building Python $PYTHON_VERSION..."

# Clean up source directory for Python before extracting

rm -rf "Python-$PYTHON_VERSION"

tar -xf "Python-$PYTHON_VERSION.tgz" || { echo "Error: Failed to extract Python source. Check $LOG_FILE for details."; exit 1; }

cd "Python-$PYTHON_VERSION" || { echo "Error: Failed to change to Python source directory. Check $LOG_FILE for details."; exit 1; }

# Configure Python to use the custom OpenSSL build and enable shared libraries.

# LD_RUN_PATH embeds runtime library paths directly into the executables.

# LDFLAGS and CPPFLAGS are passed directly to configure to ensure they are used.

LDFLAGS="-Wl,-rpath=${OPENSSL_PREFIX}/lib -Wl,-rpath=${PYTHON_PREFIX}/lib ${LDFLAGS}" \

CPPFLAGS="${CPPFLAGS}" \

./configure --prefix="$PYTHON_PREFIX" \

--with-openssl="$OPENSSL_PREFIX" \

--enable-shared \

--enable-loadable-sqlite-extensions \

LD_RUN_PATH="${OPENSSL_PREFIX}/lib:${PYTHON_PREFIX}/lib" || { echo "Error: Python configure failed. Check $LOG_FILE for details."; exit 1; }

make clean

# Explicitly set LD_LIBRARY_PATH and PYTHONPATH for the 'make' command itself.

# This ensures the newly built python executable can find its shared libraries and

# standard library modules during internal tests (like sysconfig).

echo "Running make with explicit LD_LIBRARY_PATH and PYTHONPATH..."

LD_LIBRARY_PATH="$PWD:${OPENSSL_PREFIX}/lib:${LD_LIBRARY_PATH}" \

PYTHONPATH="$SRC_DIR/Python-$PYTHON_VERSION/Lib" \

make -j"$(nproc)" || { echo "Error: Python make failed. Check $LOG_FILE for details."; exit 1; }

echo "Running make altinstall..."

make altinstall || { echo "Error: Python make altinstall failed. Check $LOG_FILE for details."; exit 1; }

echo "Python $PYTHON_VERSION installed to $PYTHON_PREFIX."

cd .. # Return to $SRC_DIR

fi

# === CREATE PYTHON VIRTUAL ENVIRONMENT FOR WHISPER ===

echo "Checking for Whisper virtual environment..."

if [ ! -d "$WHISPER_ENV" ]; then

echo "Creating Whisper virtual environment at $WHISPER_ENV."

"$PYTHON_PREFIX/bin/python3.10" -m venv "$WHISPER_ENV" || { echo "Error: Failed to create virtual environment. Check $LOG_FILE for details."; exit 1; }

else

echo "Whisper virtual environment already exists at $WHISPER_ENV."

fi

# === INSTALL WHISPER AND DEPENDENCIES ===

echo "Activating virtual environment and installing Whisper and its dependencies..."

source "$WHISPER_ENV/bin/activate" || { echo "Error: Failed to activate virtual environment. Check $LOG_FILE for details."; exit 1; }

# Upgrade pip and related tools

pip install --upgrade pip setuptools wheel || { echo "Error: Failed to upgrade pip tools. Check $LOG_FILE for details."; exit 1; }

# Install certifi explicitly to ensure it's available for SSL_CERT_FILE

echo "Installing/ensuring certifi is present in virtual environment..."

pip install certifi || { echo "Error: Failed to install certifi. Check $LOG_FILE for details."; exit 1; }

# Install Whisper from GitHub

echo "Installing OpenAI Whisper..."

pip install git+https://github.com/openai/whisper.git || { echo "Error: Failed to install Whisper. Check $LOG_FILE for details."; exit 1; }

deactivate # Deactivate the virtual environment

echo "Whisper and its dependencies installed successfully."

# === SERVICE SETUP ===

echo "--- Setting up Systemd Service for Transcriber ---"

# Enable debugging for this section to trace variable issues

set -x

# Define service-specific variables here, closer to their usage

SERVICE_NAME="transcriber"

INSTALL_DIR="/var/transcripts" # Directory for output transcripts and watcher script

SCRIPT_PATH="$INSTALL_DIR/transcribe_watcher.sh" # Watcher script path

PYTHON_BIN="$WHISPER_ENV/bin/python3" # Python executable within the virtual environment

WHISPER_BIN="$WHISPER_ENV/bin/whisper" # Whisper executable within the virtual environment

MONITOR_DIR="/var/spool/asterisk/monitor" # Directory to monitor for WAV files

# Ensure transcript directory exists

mkdir -p "$INSTALL_DIR" || { echo "Error: Failed to create install directory $INSTALL_DIR. Check $LOG_FILE for details."; exit 1; }

# Verify Python executable path for the service

if ! [ -x "$PYTHON_BIN" ]; then

echo "Error: Python executable not found or not executable at $PYTHON_BIN. Check $LOG_FILE for details."

exit 1

fi

# Verify Whisper executable path for the service

if ! [ -x "$WHISPER_BIN" ]; then

echo "Error: Whisper executable not found or not executable at $WHISPER_BIN. Check $LOG_FILE for details."

exit 1

fi

# --- START: Generate transcribe_watcher.sh directly ---

echo "Generating transcribe_watcher.sh script..."

cat > "$SCRIPT_PATH" <<'EOF'

#!/bin/bash

# This script monitors the Asterisk monitor directory for new WAV files,

# transcribes them using Whisper, and saves all output formats.

# --- Configuration ---

MONITOR_DIR="/var/spool/asterisk/monitor"

TRANSCRIPT_DIR="/var/transcripts"

WHISPER_BIN="/opt/whisper_env/bin/whisper" # Absolute path to Whisper executable

LOG_FILE="/var/log/transcriber_watcher.log" # Dedicated log for the watcher script

MIN_FILE_SIZE_KB=5 # Minimum file size in KB to process (e.g., 5KB to avoid empty recordings)

# Ensure log file exists and is writable

mkdir -p "$(dirname "$LOG_FILE")"

touch "$LOG_FILE"

exec > >(tee -a "$LOG_FILE") 2>&1

echo "Watcher script started at $(date)"

# Ensure output directory exists

mkdir -p "$TRANSCRIPT_DIR"

# --- Main Logic ---

echo "Monitoring $MONITOR_DIR for new WAV files..."

# Use inotifywait to monitor for new WAV files (close_write event)

# Process files as they are written

inotifywait -m -r "$MONITOR_DIR" -e close_write --format '%w%f' | grep --line-buffered '\.wav$' | while read -r WAV_FILE; do

echo "[$(date)] Detected new WAV file: $WAV_FILE"

# Get file size in KB

FILE_SIZE_KB=$(du -k "$WAV_FILE" | awk '{print $1}')

if (( FILE_SIZE_KB < MIN_FILE_SIZE_KB )); then

echo "[$(date)] Skipping small file (less than ${MIN_FILE_SIZE_KB}KB): $WAV_FILE"

continue

fi

# Extract filename without extension for output

FILENAME=$(basename "$WAV_FILE" .wav)

# Define output base path

OUTPUT_BASE_PATH="$TRANSCRIPT_DIR/$FILENAME"

echo "[$(date)] Transcribing: $WAV_FILE"

# Run Whisper. Use --output_format all to get srt, txt, json, tsv, vtt

# Redirect Whisper's output to a temporary file to prevent cluttering stdout/log directly

# and to capture any specific Whisper errors.

TEMP_WHISPER_LOG="/tmp/whisper_output_${FILENAME}.log"

if "$WHISPER_BIN" "$WAV_FILE" \

--model base \

--language en \

--output_dir "$TRANSCRIPT_DIR" \

--output_format all \

--verbose False > "$TEMP_WHISPER_LOG" 2>&1; then

echo "[$(date)] Transcription successful for $WAV_FILE. Outputs in $TRANSCRIPT_DIR."

# Verify if expected files exist

if [ -f "${OUTPUT_BASE_PATH}.srt" ] && [ -f "${OUTPUT_BASE_PATH}.txt" ]; then

echo "[$(date)] SRT and TXT files confirmed for $FILENAME."

# Optionally, move the processed WAV file to an archive or delete it

# For now, we will leave the original WAV file as requested by "keep all the files"

# If you want to move it: mv "$WAV_FILE" "$TRANSCRIPT_DIR/processed_wavs/"

# If you want to delete it: rm "$WAV_FILE"

else

echo "[$(date)] WARNING: Expected transcript files (.srt, .txt) not found for $FILENAME. Check $TEMP_WHISPER_LOG for Whisper output."

fi

else

echo "[$(date)] ERROR: Transcription failed for $WAV_FILE. Check $TEMP_WHISPER_LOG for Whisper output."

fi

# Clean up the temporary whisper log

rm -f "$TEMP_WHISPER_LOG"

done

EOF

echo "transcribe_watcher.sh generated and made executable at $SCRIPT_PATH."

chmod +x "$SCRIPT_PATH" || { echo "Error: Failed to make watcher script executable after generation. Check $LOG_FILE for details."; exit 1; }

# --- END: Generate transcribe_watcher.sh directly ---

# Determine the certifi CA bundle path within the virtual environment

# This path will be used to set SSL_CERT_FILE for the systemd service

echo "Determining certifi CA bundle path for SSL_CERT_FILE..."

# Use the Python executable from the virtual environment to find certifi

CERTIFI_CA_BUNDLE_PATH="$("$WHISPER_ENV/bin/python3" -c 'import certifi; print(certifi.where())')"

if [ -z "$CERTIFI_CA_BUNDLE_PATH" ]; then

echo "Error: Could not determine certifi CA bundle path. SSL_CERT_FILE might not be set correctly. Check $LOG_FILE for details."

# Optionally, you could exit here or try a fallback if absolutely necessary

else

echo "Certifi CA bundle path: $CERTIFI_CA_BUNDLE_PATH"

fi

# Create the systemd service unit file

SERVICE_FILE="/etc/systemd/system/${SERVICE_NAME}.service"

echo "Creating systemd service file: $SERVICE_FILE"

cat > "$SERVICE_FILE" <<EOF

[Unit]

Description=Real-time Call Transcriber (Whisper)

After=network.target

[Service]

Type=simple

# Corrected ExecStart to use bash to run the watcher script

ExecStart=/bin/bash $SCRIPT_PATH

WorkingDirectory=$INSTALL_DIR

Restart=always

# Ensure the virtual environment's bin directory is in PATH

Environment=PATH=$WHISPER_ENV/bin:/usr/bin:/bin

# Set SSL_CERT_FILE to the certifi bundle for secure connections

Environment=SSL_CERT_FILE=$CERTIFI_CA_BUNDLE_PATH

User=root

Group=root

[Install]

WantedBy=multi-user.target

EOF

echo "Systemd service file created."

# Reload systemd daemon, enable, and start the service

echo "Reloading systemd daemon, enabling and starting service..."

systemctl daemon-reload || { echo "Error: systemctl daemon-reload failed. Check $LOG_FILE for details."; exit 1; }

systemctl stop "$SERVICE_NAME" || { echo "Warning: Failed to stop existing service, proceeding with start. Check $LOG_FILE for details."; } # Stop before starting for clean restart

systemctl reset-failed "$SERVICE_NAME" || { echo "Warning: Failed to reset failed state for service. Check $LOG_FILE for details."; } # Clear start-limit

sleep 1 # Give systemd a moment

systemctl start "$SERVICE_NAME" || { echo "Error: systemctl start failed. Check $LOG_FILE for details."; exit 1; }

echo "Service installed and started: $SERVICE_NAME"

systemctl status "$SERVICE_NAME" --no-pager || { echo "Error: Failed to get service status. Check $LOG_FILE for details."; exit 1; }

# Disable debugging

set +x

echo "--- Whisper Installation and Service Setup Complete ---"

echo "Finished at: $(date)"

Service (transcribe_watcher.sh)

#!/bin/bash

# This script monitors the Asterisk monitor directory for new WAV files,

# transcribes them using Whisper, and saves all output formats.

# --- Configuration ---

MONITOR_DIR="/var/spool/asterisk/monitor"

TRANSCRIPT_DIR="/var/transcripts"

WHISPER_BIN="/opt/whisper_env/bin/whisper" # Absolute path to Whisper executable

LOG_FILE="/var/log/transcriber_watcher.log" # Dedicated log for the watcher script

MIN_FILE_SIZE_KB=5 # Minimum file size in KB to process (e.g., 5KB to avoid empty recordings)

# Ensure log file exists and is writable

mkdir -p "$(dirname "$LOG_FILE")"

touch "$LOG_FILE"

exec > >(tee -a "$LOG_FILE") 2>&1

echo "Watcher script started at $(date)"

# Ensure output directory exists

mkdir -p "$TRANSCRIPT_DIR"

# --- Main Logic ---

echo "Monitoring $MONITOR_DIR for new WAV files..."

# Use inotifywait to monitor for new WAV files (close_write event)

# Process files as they are written

inotifywait -m -r "$MONITOR_DIR" -e close_write --format '%w%f' | grep --line-buffered '\.wav$' | while read -r WAV_FILE; do

echo "[$(date)] Detected new WAV file: $WAV_FILE"

# Get file size in KB

FILE_SIZE_KB=$(du -k "$WAV_FILE" | awk '{print $1}')

if (( FILE_SIZE_KB < MIN_FILE_SIZE_KB )); then

echo "[$(date)] Skipping small file (less than ${MIN_FILE_SIZE_KB}KB): $WAV_FILE"

continue

fi

# Extract filename without extension for output

FILENAME=$(basename "$WAV_FILE" .wav)

# Define output base path

OUTPUT_BASE_PATH="$TRANSCRIPT_DIR/$FILENAME"

echo "[$(date)] Transcribing: $WAV_FILE"

# Run Whisper. Use --output_format all to get srt, txt, json, tsv, vtt

# Redirect Whisper's output to a temporary file to prevent cluttering stdout/log directly

# and to capture any specific Whisper errors.

TEMP_WHISPER_LOG="/tmp/whisper_output_${FILENAME}.log"

if "$WHISPER_BIN" "$WAV_FILE" \

--model base \

--language en \

--output_dir "$TRANSCRIPT_DIR" \

--output_format all \

--verbose False > "$TEMP_WHISPER_LOG" 2>&1; then

echo "[$(date)] Transcription successful for $WAV_FILE. Outputs in $TRANSCRIPT_DIR."

# Verify if expected files exist

if [ -f "${OUTPUT_BASE_PATH}.srt" ] && [ -f "${OUTPUT_BASE_PATH}.txt" ]; then

echo "[$(date)] SRT and TXT files confirmed for $FILENAME."

# Optionally, move the processed WAV file to an archive or delete it

# For now, we will leave the original WAV file as requested by "keep all the files"

# If you want to move it: mv "$WAV_FILE" "$TRANSCRIPT_DIR/processed_wavs/"

# If you want to delete it: rm "$WAV_FILE"

else

echo "[$(date)] WARNING: Expected transcript files (.srt, .txt) not found for $FILENAME. Check $TEMP_WHISPER_LOG for Whisper output."

fi

else

echo "[$(date)] ERROR: Transcription failed for $WAV_FILE. Check $TEMP_WHISPER_LOG for Whisper output."

fi

# Clean up the temporary whisper log

rm -f "$TEMP_WHISPER_LOG"

done