expressive-tts

Expressive TTS (Lightning v3.2)

Control emotion, pitch, volume, speaking rate, and accent — make the same voice sound happy, angry, whispering, sarcastic, or anything in between.

Note: v3.2 is currently available on waves-api.smallest.ai only.

Try It Now

curl -o happy.wav \
  -X POST "https://waves-api.smallest.ai/api/v1/lightning-v3.2/get_speech" \
  -H "Authorization: Bearer $SMALLEST_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "text": "This is absolutely incredible! I cannot believe how amazing this sounds!",
    "voice_id": "natalie",
    "emotion": "excited",
    "pitch": "high-pitched",
    "volume": "normal",
    "prosody": "fast",
    "accent": "general american",
    "sample_rate": 44100,
    "output_format": "wav"
  }'

Features

• 19 emotions: happy, sad, angry, excited, calm, sarcastic, frustrated, fearful, surprised, disgusted, bored, anxious, confident, amused, empathetic, nostalgic, pleading, skeptical, neutral
• 4 pitch styles: mid-range, high-pitched, low-pitched, breathy
• 6 volume levels: normal, shouting, soft, whispering, muttering, loud
• 9 speaking rates: normal, very slow, slow, fast, very fast, melodic, monotonous, hesitant, measured
• 9 accents: general american, british, australian, indian american, scottish, irish, southern american, new york, canadian

Usage

Hardcoded Emotions

export SMALLEST_API_KEY="your-key"

# Generate 6 different emotional styles
python expressive.py

# Or specific emotion
python expressive.py --emotion angry --accent british --text "This is unacceptable!"

Auto-Detect Emotion with LLM

The LLM reads the text and predicts the best emotion, pitch, volume, prosody, and accent automatically:

export SMALLEST_API_KEY="your-key"
export OPENAI_API_KEY="your-openai-key"

python llm_predict_and_speak.py "WHAT DID YOU JUST SAY TO ME?!"
# → Predicts: angry, high-pitched, shouting, fast
# → Generates angry_shouting_fast.wav

WebSocket Streaming

python stream_expressive.py "Take a deep breath. Everything is going to be fine." --emotion calm --volume soft

Important: Sample Rate

v3.2 outputs audio at 44100 Hz (not 24000 like v3.1). Using 24000 will make audio sound muffled.

What's Next?

Want to…	Go to
Use standard TTS (v3.1)	Getting Started
Build a voice game with emotions	Chinese Whispers Game
Browse all voices	Voices