Text-to-Speech

The agent's voice. Modern TTS can stream audio in real-time, sound nearly human, and even clone specific voices. Here's your toolkit.

Local / Open-Source Tools

Run text-to-speech entirely on your machine — no cloud, no API keys, no audio leaving your network. These tools let you synthesize speech from text, compare voice quality across engines, and even clone voices locally.

Piper TTS

ONNX-based neural TTS that runs on minimal hardware — even a Raspberry Pi. Pre-installed and available system-wide. Fast synthesis but limited to sentence-level streaming.

bash

# Piper is pre-installed and available system-wide
# Wrapper at /usr/local/bin/piper → /opt/piper/speak.py

# Basic text-to-speech
echo "Hello, this is Piper text to speech." \
  | piper --model /opt/piper/models/en_US-lessac-medium.onnx --output_raw \
  | aplay -r 22050 -f S16_LE

# Save to a file instead of playing
echo "Save this to a file." \
  | piper --model /opt/piper/models/en_US-lessac-medium.onnx --output_file output.wav

Kokoro TTS

82M parameter TTS model with excellent quality and natural prosody. ONNX-based, runs on CPU. Higher quality than Piper but slightly slower. Streaming server available via Kokoro-FastAPI.

bash

# Kokoro is pre-installed and available system-wide
# Wrapper at /usr/local/bin/kokoro → /opt/kokoro/speak.py

# Basic text-to-speech
echo "Hello, this is Kokoro text to speech." \
  | kokoro \
  | aplay -r 24000 -f S16_LE

# Kokoro-FastAPI streaming server (if running)
# Web UI at http://localhost:8880/web
time curl -s http://localhost:8880/v1/audio/speech \
  -H "Content-Type: application/json" \
  -d '{"input": "Hello from Kokoro streaming server.", "voice": "af_heart", "response_format": "wav"}' \
  -o output.wav && aplay output.wav

Coqui XTTS

Zero-shot voice cloning and TTS — clone a voice from just 6 seconds of audio. Requires more compute than Piper or Kokoro. We use the original TTS package, not the community coqui-tts fork.

Coqui AI shut down in early 2024. This package has known dependency issues and is not actively maintained. Use for demos and experimentation only — not recommended for production.

bash

# Coqui TTS is pre-installed and available system-wide
# Wrapper at /usr/local/bin/coqui-tts → /opt/coqui-tts/.venv/bin/tts

# Basic text-to-speech
coqui-tts --text "Hello, this is Coqui TTS." --out_path output.wav
aplay output.wav

# List available models
coqui-tts --list_models

# Voice cloning with XTTS v2 (provide a reference audio clip)
coqui-tts --model_name tts_models/multilingual/multi-dataset/xtts_v2 \
  --text "Hello, this is a cloned voice." \
  --speaker_wav reference_voice.wav \
  --language_idx en \
  --out_path cloned_output.wav

Commercial Cloud Alternatives

These companies are not sponsors or affiliated with this training or Call Center Village. They're listed for educational awareness only.
That being said, if any of you are reading this — Call Center Village is always looking for sponsors!

ElevenLabs ↗

Industry-leading voice quality with streaming API and voice cloning. Used by many production voice agents.

Cartesia (Sonic) ↗

Built specifically for real-time voice agents. State-space architecture with ultra-low latency.

Murf AI ↗

AI voice platform with text-to-speech, voice cloning, and AI dubbing tools.

Deepgram Aura ↗

TTS optimized for voice agents with extremely low latency. From the same company as their STT product.

Have a suggestion for a TTS service to include here? Email us at [email protected]

Choosing TTS for Your Agent

The right TTS depends on your use case — speed, quality, privacy, and budget all factor in.

Scenario	Recommended TTS	Why
Quick local demo	Piper	Fastest on CPU, minimal setup
High-quality local demo	Kokoro	Best quality for local, runs on CPU
Voice cloning attack sim	Coqui XTTS	Local cloning + generation in one
Edge / IoT deployment	Piper	Runs on minimal hardware
Production voice agent	ElevenLabs / Cartesia	Best quality + streaming + low latency

Check Your Understanding

Question 1 / 3

What is the main advantage of Piper TTS over Kokoro TTS?

Loading…