VoxCPM 1.5 is an end-to-end text-to-speech (TTS) model from ModelBest. It features zero-shot voice cloning and high-quality speech synthesis capabilities.
Links
Tags
NeuTTS Air is the world's first super-realistic, on-device TTS speech language model with instant voice cloning. Built on a 0.5B LLM backbone, it brings natural-sounding speech, real-time performance, and speaker cloning to local devices.
Links
Tags
Qwen3-TTS-12Hz-1.7B-CustomVoice via vLLM-Omni - Text-to-speech model from Alibaba Qwen team with custom voice cloning capabilities. Generates natural-sounding speech with voice personalization.
Links
Tags
Qwen3-TTS 0.6B Base (C++ / GGML, qwentts.cpp). Native C++ text-to-speech with streaming output and zero-shot voice cloning (set `voice` to a 24kHz reference .wav). 24kHz mono, 11 languages with Mandarin dialects. Q8_0 (~0.95 GB talker).
Links
Tags
Qwen3-TTS 0.6B Base (C++ / GGML, qwentts.cpp), Q4_K_M (~0.6 GB talker). Streaming + voice cloning, 24kHz mono, 11 languages.
Links
Tags
Qwen3-TTS 1.7B Base (C++ / GGML, qwentts.cpp), Q8_0 (~2.0 GB talker). Higher-quality streaming + voice cloning, 24kHz mono, 11 languages.
Links
Tags
Qwen3-TTS 1.7B Base (C++ / GGML, qwentts.cpp), Q4_K_M (~1.2 GB talker). Streaming + voice cloning, 24kHz mono, 11 languages.
Links
Tags
Repository: localaiLicense: apache-2.0
MOSS-TTS-Local v1.5 (C++/ggml, moss-tts.cpp), Q8_0 (~6 GB), near-lossless. Native C++ text-to-speech for the OpenMOSS MOSS-TTS-Local Transformer v1.5 (a GPT-J local transformer decoded through the MOSS-Audio-Tokenizer-v2 neural codec), 48kHz stereo output with reference-audio voice cloning (set `voice` to a reference .wav). Verified per component against the reference PyTorch and about 2x faster per frame on CPU.
Links
Tags
MOSS-TTS-Local v1.5 (C++/ggml, moss-tts.cpp), F16 (~9.4 GB), code-exact versus the reference. 48kHz stereo text-to-speech with reference-audio voice cloning.
Links
Tags
OmniVoice (C++ / GGML) - native text-to-speech with voice cloning and voice design. 24kHz mono output, 646 languages, streaming synthesis. Q8_0 GGUFs (~945 MB total): 612M Qwen3 backbone + RVQ audio codec.
Links
Tags
OmniVoice (C++ / GGML), BF16 high-quality variant - text-to-speech with voice cloning and voice design. 24kHz mono, 646 languages, streaming. BF16 GGUFs (~1.6 GB total).
Links
Tags

Fish Speech S2-Pro is a high-quality text-to-speech model supporting voice cloning via reference audio. Uses a two-stage pipeline: text to semantic tokens (LLaMA-based) then semantic to audio (DAC decoder).
Links
Tags
Repository: localaiLicense: mit

Chatterbox, Resemble AI's first production-grade open source TTS model. Supports zero-shot voice cloning from a reference WAV, including saved LocalAI Voice Library profiles. Licensed under MIT, Chatterbox has been benchmarked against leading closed-source systems like ElevenLabs, and is consistently preferred in side-by-side evaluations.
Links
Tags
Repository: localai
F5-TTS v1 Base (SWivid, MIT) text-to-speech synthesized through the CrispASR backend. A 22-layer DiT flow-matching model with a built-in Vocos vocoder in a single self-contained GGUF (no separate codec). Auto-detected by CrispASR and runs end-to-end on CPU, producing 24 kHz mono audio. F5-TTS is a voice-cloning model with no built-in speaker: you must supply a reference clip and its exact transcript. Select a saved LocalAI Voice Library profile per request, or configure `voice:` plus `voice_text:` in the model's `options` as a model-wide fallback; without a reference the model cannot generate audio. Default GGUF size ~945 MB (f16). Quantization below f16 is not recommended for flow-matching models. Synthesis runs a 32-step ODE solver and is compute-heavy on CPU (expect long generation times without a GPU-enabled CrispASR build).
Links
Tags
Repository: localaiLicense: mit
Chatterbox (audio.cpp, Q8_0) - multilingual zero-shot voice cloning, served by the audio-cpp backend. A reference clip is REQUIRED. This build advertises only the cloning and voice-conversion tasks, not plain TTS, so every request must set the TTS voice field to the path of a WAV file on the server; a voice that is not a file on disk is treated as a named preset and the request is refused with a message saying so. The voice-conversion half is reached through /audio/transform rather than /v1/audio/speech: post the speech to convert as `audio` and the target speaker's clip as `reference`. That is why this entry declares audio_transform alongside tts. With both this and a source-separation model installed, /audio/transform has two candidates, so name the model explicitly rather than relying on the default pick. Q8_0 rather than 16-bit because upstream's tested-status table records the same result for both (Pass, ASR match with drift) and this build is half the size.
Links
Tags