Transcribe Audio

Transcribe audio files to text using the local Whisper ASR service running at http://localhost:9000.

Prerequisites

• The Whisper service must be running (check with curl http://localhost:9000/health)
• Audio file must be in a supported format: OGG, MP3, M4A, WAV, or WebM
• Audio must be in English (the local model is English-only)

Usage

Basic Transcription

To transcribe an audio file:

curl -X POST http://localhost:9000/asr \
  -F "audio_file=@/path/to/audio.wav" \
  -F "language=en"

Save to File

Save the transcription to the agent's local folder:

curl -X POST http://localhost:9000/asr \
  -F "audio_file=@/path/to/audio.wav" \
  -F "language=en" \
  -o transcription.txt

Or with a specific output path:

OUTPUT_FILE="${PWD}/$(basename /path/to/audio.wav .wav).txt"
curl -X POST http://localhost:9000/asr \
  -F "audio_file=@/path/to/audio.wav" \
  -F "language=en" \
  -o "$OUTPUT_FILE"

Quick Workflow

1. Verify the service is running:

   curl -s http://localhost:9000/health

   Expected response: {"status": "ok"}
2. Transcribe and save:

   curl -X POST http://localhost:9000/asr \
     -F "audio_file=@<audio-file-path>" \
     -F "language=en" \
     -o <output-file-path>

3. Verify the output:

   cat <output-file-path>

Supported Audio Formats

Performance Notes

• The local Whisper service uses the tiny.en model
• Processing time is roughly 10-20% of audio duration (e.g., 1 minute of audio takes ~6-12 seconds)
• The service only supports English transcription
• For best results, ensure audio has clear speech with minimal background noise

Error Handling

Common issues and solutions:

Example: Transcribe a Single File

# Set variables
AUDIO_FILE="/path/to/recording.wav"
OUTPUT_FILE="${PWD}/transcription.txt"

# Verify service
if ! curl -s http://localhost:9000/health | grep -q '"status": "ok"'; then
    echo "Error: Whisper service is not running"
    exit 1
fi

# Transcribe
curl -X POST http://localhost:9000/asr \
  -F "audio_file=@${AUDIO_FILE}" \
  -F "language=en" \
  -o "$OUTPUT_FILE"

echo "Transcription saved to: $OUTPUT_FILE"
cat "$OUTPUT_FILE"

Example: Batch Transcribe Multiple Files

for audio_file in *.wav; do
    output_file="${PWD}/$(basename "$audio_file" .wav).txt"
    echo "Transcribing: $audio_file -> $output_file"
    curl -X POST http://localhost:9000/asr \
      -F "audio_file=@${audio_file}" \
      -F "language=en" \
      -o "$output_file"
done