Lmnt Ai

LMNT specializes in ultra-low-latency TTS for conversational AI: its streaming API delivers the first audio byte in under 100ms, making it suitable for real-time voice agent pipelines where latency matters (phone bots, live assistants, interactive storytelling). Offers voice cloning from short audio samples for personalized voice synthesis. For agents: synthesize speech as a stream (play while still generating), use in conjunction with STT + LLM for full voice pipeline, or generate audio files for async delivery. Competitive with ElevenLabs on voice quality; differentiated by latency. Confidence is docs-derived.

REST API at api.lmnt.com/v1. Resources: speech, voice. POST /v1/ai/speech/bytes generates speech and returns audio bytes (synchronous; for short texts). POST /v1/ai/speech/stream generates speech as a streaming response (chunked transfer encoding; audio bytes stream as generated). Request body: {text, voice, format: mp3/wav/aac, sample_rate, speed, seed}. GET /v1/ai/voice lists available voices (built-in + cloned). POST /v1/ai/voice creates a voice clone (multipart upload of audio sample, name). DELETE /v1/ai/voice/{id} removes a cloned voice. Response: audio bytes (speech endpoints); JSON (voice management).

API key auth: X-API-Key: {api_key} header. Keys from LMNT Account Settings → API Keys. Account-scoped. HTTPS enforced. No OAuth2. Cloned voices are account-scoped — not shareable across accounts. Rate limiting per API key (plan-dependent). Audio output is not stored server-side — received directly in response body.

Standard HTTP status codes with JSON error bodies. 401 for invalid API key, 400 for invalid voice ID or unsupported format, 422 for text too long (per-request text length limits apply). Streaming: chunked transfer — handle partial audio chunks in client; retry on connection drop. Audio quality depends on voice selection and input text. Voice cloning quality depends on input sample quality (clear speech, minimal background noise). LMNT uptime at status.lmnt.com. Rate limits enforced at request level — queue if concurrency exceeds plan limits.

docs.lmnt.com provides REST API reference with request/response examples, voice cloning guide, and streaming integration walkthrough. Python SDK (lmnt-python) for idiomatic use. Getting started: create account, get API key, first speech synthesis in under 5 minutes. Free tier available. Limited but focused documentation — LMNT's product is narrowly scoped (TTS + cloning), so docs are concise. Community support via LMNT support channels. Growing ecosystem in the voice AI space given the latency advantage.