Features
Realtime streaming TTS with sub-200ms first-chunk latency
Instant voice cloning from 5–15 seconds audio
Text-based voice design via natural language description
Cross-lingual cloning: one voice speaks 100+ languages natively
Voice steering via bracketed instructions (tone, speed, volume, style)
Multilingual support: 100+ languages (TTS-2), 15 languages (TTS 1.5)
Multiple model tiers: Realtime TTS-2, TTS 1.5 Max, TTS 1.5 Mini
Word-level timestamp alignment
WebSocket support for realtime audio streaming
Unified API for TTS, STT, and LLM routing (220+ models)
Custom pronunciation dictionary
Speaking rate and temperature control
Multiple audio encodings: OGG_OPUS, WAV, PCM, etc.
TTS Playground for testing voices before API integration
Professional voice cloning (add-on for higher tiers)
Text-to-speech (Octave 2) with emotional control across 48 emotions and 600+ voice descriptors
Speech-to-speech (EVI 4 mini) with interruptibility, back-channeling, and configurable turn detection
Real-time configurable turn detection (end_of_turn_silence_ms, speech_detection_threshold)
Real-time configurable interruption settings (min_interruption_ms)
Zero prompt expansion for external LLMs (GPT-5.2, Claude Opus 4.6, etc.)
Experimental temperature parameter for TTS sampling variation
Voice cloning (unlimited create/use, API access on Business tier)
Voice conversion with commercial license on Creator+
50+ languages, 48+ emotions, 600+ voice descriptors
Human Feedback API with science-backed survey templates
Curated speech datasets for training (multilingual, emotional, domain-specific)
SOC 2 Type II, GDPR, HIPAA compliance (Business and Enterprise tiers)
RESTful API for evaluation studies (coming soon)
Open-source models (TADA) available separately
Concurrent connections for external LLMs (up to 30 on Business)
