Unreal Speech

About Unreal Speech

Unreal Speech is a text-to-speech API that competes almost entirely on cost, claiming to be roughly eleven times cheaper than ElevenLabs and putting a side-by-side monthly comparison on its own homepage. It runs on Kokoro-82M, a small open-weights TTS model, which is what makes the price structure possible. The technical specifications are aimed at production rather than demos: audio starts streaming in about 300 milliseconds, a single request can produce up to ten hours of audio, and per-word timestamps come back with the synthesis so you can highlight words in sync with playback. Timestamps are available in three ways — a websocket endpoint that streams audio and timestamps together, or the /speech and /synthesisTasks endpoints with TimestampType set to word or sentence, which return a JSON URI of word/start/end/text_offset objects. The long-request ceiling matters for audiobook and long-form article narration, which is exactly where per-character billing on premium vendors becomes painful. There is a browser Studio for non-API use, a live demo with adjustable voice, language, speed, pitch, and bitrate, and a free tier of 250,000 characters — around six hours of audio, roughly half an audiobook — which is large enough to actually evaluate the quality on real content rather than on a sample paragraph. The team is based in San Francisco and publishes a technical blog on open TTS models.

Key Features

Unreal Speech Pros & Cons

Who Is Unreal Speech Best For?

Tags