Models and engines

Small, fast TTS with a curated voice pack. Runs in real time on CPU.

Zero-shot voice cloning with an emotion exaggeration control and built-in watermarking.

Multilingual voice cloning from a six second reference clip. Seventeen languages.

Flow-matching TTS with strong zero-shot cloning and natural prosody.

Llama-based speech LLM with paralinguistic tags such as laughs and sighs.

Streaming zero-shot TTS with instruction control over style and dialect.

Dialogue generation with speaker tags and nonverbal cues in a single pass.

Multilingual TTS with few-shot cloning across thirteen languages.

Voice design from a text description: gender, pace, pitch, recording quality.

Generative audio model that produces speech, music snippets and sound effects.

Very fast ONNX voices designed for low-power devices. Large curated voice set.

CPU real-time multilingual TTS with mixed Chinese and English support.

Conversational speech model conditioned on prior dialogue turns.

Expressive TTS with explicit controls for pitch, speaking rate and emotion vectors.

Lightweight encoder-decoder TTS using x-vector speaker embeddings.

Fifteen million parameter TTS that runs anywhere a CPU does.

Tokenizer-free TTS with context-aware prosody and voice cloning.

Duration-controllable zero-shot TTS with disentangled timbre and emotion.

Audio foundation model for expressive multi-speaker generation.

Reference multilingual transcription and translation model.

Pruned decoder variant of large-v3 with several times faster decoding.

CTranslate2 port of Whisper with int8 quantization and word timestamps.

Distilled English Whisper. Roughly six times faster with near-identical WER.

FastConformer TDT model with punctuation and word-level timestamps.

Multitask ASR and speech translation across four languages.

Edge-focused ASR whose compute scales with input length. Good for dictation.

Speech-language model for transcription plus audio question answering.

Streaming speech-to-text with semantic voice activity detection.

ASR with emotion and audio event tags in the transcript.

CTC baseline for English. Useful for forced alignment and quick checks.

Zero-shot voice conversion and singing voice conversion.

Two-stage denoiser and enhancer for cleaning up reference clips before cloning.

Route speech generation to any server exposing /v1/audio/speech.

Route transcription to any server exposing /v1/audio/transcriptions.