Skip to main content

Audio Commands

The lc tool provides comprehensive audio support, enabling both speech-to-text (transcription) and text-to-speech (TTS) capabilities.

Audio Transcription​

Convert audio files to text using the transcribe command.

Basic Usage​

# Transcribe a single audio file
lc transcribe audio.wav

# Transcribe multiple audio files
lc transcribe file1.mp3 file2.wav file3.flac

# Use alias
lc tr recording.mp3

Options​

  • -m, --model <MODEL> - Specify the transcription model (default: whisper-1)
  • -p, --provider <PROVIDER> - Specify the provider (default: openai)
  • -l, --language <LANG> - Specify the audio language (e.g., en, es, fr)
  • --prompt <TEXT> - Provide context to guide the transcription
  • -f, --format <FORMAT> - Output format: text, json, srt, vtt (default: text)
  • -t, --temperature <TEMP> - Sampling temperature (0-1, default: 0)

Supported Audio Formats​

  • MP3, MP4, MPEG, MPGA, M4A
  • WAV
  • WebM
  • OGG
  • FLAC

Examples​

# Transcribe with language hint
lc transcribe interview.mp3 --language en

# Get JSON output with timestamps
lc transcribe podcast.wav --format json

# Provide context for better accuracy
lc transcribe medical_recording.mp3 --prompt "Medical consultation discussing symptoms"

# Use a different provider
lc transcribe audio.wav --provider custom-provider

Text-to-Speech (TTS)​

Convert text to speech using the tts command.

Basic Usage​

# Generate speech from text
lc tts "Hello, this is a test of text to speech"

# Save to a specific file
lc tts "Welcome to our service" --output welcome.mp3

# Read from a file
lc tts --file script.txt --output narration.mp3

Options​

  • -m, --model <MODEL> - TTS model: tts-1, tts-1-hd (default: tts-1)
  • -p, --provider <PROVIDER> - Specify the provider (default: openai)
  • -v, --voice <VOICE> - Voice selection: alloy, echo, fable, onyx, nova, shimmer (default: alloy)
  • -o, --output <FILE> - Output file path (default: speech_TIMESTAMP.mp3)
  • -f, --format <FORMAT> - Audio format: mp3, opus, aac, flac, wav, pcm (default: mp3)
  • -s, --speed <SPEED> - Speech speed: 0.25 to 4.0 (default: 1.0)
  • --file <FILE> - Read text from a file instead of command line

Voice Options​

  • alloy - Neutral and balanced
  • echo - Warm and conversational
  • fable - Expressive and dynamic
  • onyx - Deep and authoritative
  • nova - Friendly and upbeat
  • shimmer - Soft and pleasant

Examples​

# Use HD model for better quality
lc tts "Important announcement" --model tts-1-hd

# Generate with different voice
lc tts "Welcome message" --voice nova --output welcome.mp3

# Adjust speech speed
lc tts "Quick instructions" --speed 1.5

# Generate in different format
lc tts "Audio book chapter" --format flac --output chapter1.flac

# Read from file with specific voice
lc tts --file presentation.txt --voice onyx --output presentation.mp3

Audio Attachments in Chat​

You can attach audio files to chat prompts for context-aware conversations.

Basic Usage​

# Ask about audio content
lc "What is being discussed in this recording?" --audio meeting.mp3

# Multiple audio files
lc "Summarize these interviews" --audio interview1.mp3 --audio interview2.wav

# Combine with other attachments
lc "Analyze this presentation" --audio narration.mp3 --image slides.png

How It Works​

  1. Audio files are automatically transcribed using the Whisper model
  2. Transcriptions are added to the chat context
  3. The LLM processes both your prompt and the transcribed content

Examples​

# Meeting transcription and summary
lc "Provide meeting minutes for this recording" --audio meeting_recording.mp3 -m gpt-4o

# Language translation
lc "Translate this Spanish audio to English" --audio spanish_audio.wav

# Content analysis
lc "What are the key points discussed?" --audio podcast_episode.mp3

# Multiple audio analysis
lc "Compare the topics discussed in these recordings" \
  --audio episode1.mp3 \
  --audio episode2.mp3 \
  -m claude-3-opus-20240229

Configuration​

Setting Default Audio Provider​

# Set default provider for audio commands
lc config set audio-provider openai

# Set default transcription model
lc config set audio-model whisper-1

# Set default TTS model
lc config set tts-model tts-1-hd

# Set default TTS voice
lc config set tts-voice nova

Provider Configuration​

Audio endpoints can be configured in provider TOML files:

[provider]
name = "custom-audio"
base_url = "https://api.custom.com/v1"

# Audio transcription endpoint
audio_path = "/audio/transcriptions"

# Text-to-speech endpoint  
speech_path = "/audio/speech"

# Optional: Custom templates for request/response transformation
[audio_templates]
whisper-1 = "custom_whisper_template.tera"

[speech_templates]
tts-1 = "custom_tts_template.tera"

Tips and Best Practices​

Audio Quality​

  • Use high-quality audio files for better transcription accuracy
  • WAV and FLAC formats preserve quality better than compressed formats
  • For TTS, use tts-1-hd model for production-quality output

Performance​

  • Transcription time depends on audio length and quality
  • Batch process multiple files when possible
  • Consider using JSON format for transcriptions if you need timestamps

Language Support​

  • Whisper supports 50+ languages for transcription
  • Always specify language hint for non-English audio
  • TTS voices are optimized for English but support multiple languages

Cost Optimization​

  • tts-1 is faster and cheaper than tts-1-hd
  • Compress audio files before transcription to reduce upload time
  • Use appropriate audio formats (MP3 for speech, WAV for music)

Error Handling​

Common issues and solutions:

# Unsupported format error
# Solution: Convert to supported format first
ffmpeg -i audio.aiff -acodec libmp3lame audio.mp3
lc transcribe audio.mp3

# Large file error
# Solution: Split or compress the audio
ffmpeg -i large_file.wav -t 3600 part1.wav  # First hour
lc transcribe part1.wav

# API rate limits
# Solution: Add delays between requests or batch process
lc transcribe *.mp3 --delay 1

Integration Examples​

Podcast Workflow​

# Transcribe podcast
lc transcribe podcast.mp3 --format srt --output podcast.srt

# Generate summary
lc "Summarize this podcast transcript" --file podcast.srt

# Create promotional audio
lc tts --file summary.txt --voice nova --output promo.mp3

Meeting Assistant​

# Transcribe meeting
lc transcribe meeting.m4a --output meeting.txt

# Extract action items
lc "List all action items from this meeting" --file meeting.txt

# Generate audio summary
lc tts "Meeting summary: $(cat summary.txt)" --output meeting_summary.mp3

Language Learning​

# Transcribe foreign language audio
lc transcribe spanish_lesson.mp3 --language es

# Get translation
lc "Translate this to English and explain grammar" --audio spanish_lesson.mp3

# Generate pronunciation guide
lc tts "Hola, ¿cómo estás?" --voice shimmer --speed 0.8