Audio generation

Generate speech audio using a configured text-to-speech provider.

term-llm audio "hello from term-llm"

By default, audio clips are:

• Saved to ~/Music/term-llm/ with timestamped filenames
• Generated with Venice tts-kokoro
• Rendered as MP3 using voice af_sky

You can also use Gemini or ElevenLabs TTS:

term-llm audio "Say cheerfully: hello from Gemini" --provider gemini --voice Kore
term-llm audio "Hello from ElevenLabs" --provider elevenlabs --voice Rachel --model eleven_flash_v2_5

Audio Flags

--model, --voice, --voice1, --voice2, --format, --provider, and --apply-text-normalization include shell completion candidates.

Examples

term-llm audio "hello from term-llm"
term-llm audio "quick smoke test" --output smoke.mp3
term-llm audio "faster please" --speed 1.25 --format wav
term-llm audio "sad robot noises" \
  --model tts-qwen3-0-6b \
  --voice Vivian \
  --prompt "Sad and slow."

term-llm audio "Say cheerfully: have a wonderful day" \
  --provider gemini \
  --model gemini-3.1-flash-tts-preview \
  --voice Kore \
  --format wav

term-llm audio "TTS the following conversation between Joe and Jane: Joe: Hi Jane. Jane: Hi Joe." \
  --provider gemini \
  --speaker1 Joe --voice1 Kore \
  --speaker2 Jane --voice2 Puck

term-llm audio "A one second ElevenLabs smoke test." \
  --provider elevenlabs \
  --model eleven_flash_v2_5 \
  --voice Rachel \
  --format mp3_44100_128 \
  --stability 0.5 \
  --similarity-boost 0.75

echo "pipe me" | term-llm audio --voice af_bella -o - > out.mp3

Venice TTS Models

term-llm includes the Venice text-to-speech model catalog:

Voices are model-specific. If a model rejects a voice, Venice returns the API error directly.

Gemini TTS Models

Gemini TTS returns 24 kHz mono PCM. term-llm can save it directly as pcm, or wrap it as a wav file. Gemini TTS does not support streaming; language is auto-detected.

Gemini prebuilt voices:

Zephyr, Puck, Charon, Kore, Fenrir, Leda, Orus, Aoede, Callirrhoe, Autonoe, Enceladus, Iapetus, Umbriel, Algieba, Despina, Erinome, Algenib, Rasalgethi, Laomedeia, Achernar, Alnilam, Schedar, Gacrux, Pulcherrima, Achird, Zubenelgenubi, Vindemiatrix, Sadachbia, Sadaltager, Sulafat.

ElevenLabs TTS Models

term-llm includes the documented ElevenLabs text-to-speech models:

ElevenLabs voices are account-specific. --voice accepts either a raw voice_id or an exact voice name from the account; name lookup uses the ElevenLabs voices API before generating speech.

ElevenLabs output formats:

alaw_8000, mp3_22050_32, mp3_24000_48, mp3_44100_32, mp3_44100_64, mp3_44100_96, mp3_44100_128, mp3_44100_192, opus_48000_32, opus_48000_64, opus_48000_96, opus_48000_128, opus_48000_192, pcm_8000, pcm_16000, pcm_22050, pcm_24000, pcm_32000, pcm_44100, pcm_48000, ulaw_8000, wav_8000, wav_16000, wav_22050, wav_24000, wav_32000, wav_44100, wav_48000.

JSON Output

--json prints a single structured object to stdout after saving the file.

{
  "provider": "venice",
  "text": "hello from term-llm",
  "model": "tts-kokoro",
  "voice": "af_sky",
  "format": "mp3",
  "output": {
    "path": "/home/me/Music/term-llm/20260502-120000-hello_from_term-llm.mp3",
    "mime_type": "audio/mpeg",
    "bytes": 12345
  }
}

Credentials and Config

term-llm audio reads Venice credentials from VENICE_API_KEY, audio.venice.api_key, or the existing image.venice.api_key fallback.

Gemini credentials are read from GEMINI_API_KEY, audio.gemini.api_key, image.gemini.api_key, or the configured providers.gemini API key.

ElevenLabs credentials are read from ELEVENLABS_API_KEY, XI_API_KEY, or audio.elevenlabs.api_key.

audio:
  provider: venice
  output_dir: ~/Music/term-llm
  venice:
    api_key: $VENICE_API_KEY
    model: tts-kokoro
    voice: af_sky
    format: mp3
  gemini:
    api_key: $GEMINI_API_KEY
    model: gemini-3.1-flash-tts-preview
    voice: Kore
    format: wav
  elevenlabs:
    api_key: $ELEVENLABS_API_KEY
    model: eleven_multilingual_v2
    voice: JBFqnCBsd6RMkjVDRZzb
    format: mp3_44100_128