Maya Research · Text-to-Speech

Maya1

3B-parameter Llama-style TTS that turns a plain-language voice description and inline emotion tags (laugh, cry, whisper, angry) into 24kHz streamed speech via the SNAC codec, sub-100ms latency on one 16GB+ GPU. Apache-2.0; Maya Research positions it as feature parity with ElevenLabs sans per-second fees.