Kokoro-82M | TTS (default, built-in voices; needs espeak-ng) | ~350 MB |
Qwen3-TTS-12Hz-0.6B-Base | TTS (base) | ~2.3 GB |
Qwen3-TTS-12Hz-0.6B-CustomVoice | TTS (built-in voices) | ~2.3 GB |
Qwen3-TTS-12Hz-1.7B-VoiceDesign | TTS (voice design) | ~4.2 GB |
Voxtral-4B-TTS-2603 | TTS (20 preset voices, CC BY-NC 4.0) | ~8.1 GB |
Chatterbox-500M | TTS (English, cloning from audio alone, watermarked) | ~3.2 GB |
Chatterbox-Multilingual-500M | TTS (22 languages, cloning from audio alone, watermarked) | ~3.3 GB |
Maya1-3B | TTS (voice design from a description, emotion tags) | ~6.7 GB |
Maya1-3B-GGUF | TTS (Maya1 with a Q8_0 decoder) | ~3.6 GB |
Maya2-Native-2.5B | TTS (English and 11 Indic languages, 20 voices, cloning across languages) | ~5.5 GB |
Parakeet-TDT-0.6B-v3 | ASR (default) | ~2.5 GB |
Qwen3-ASR-0.6B-GGUF | ASR (compact) | ~1.0 GB |
Nemotron-3.5-ASR-Streaming-0.6B | ASR (NVIDIA multilingual .nemo) | ~2.37 GB |
Granite-Speech-4.1-2B-Plus | ASR (IBM rich transcription) | ~4.2 GB |
Granite-Speech-5.0-2B | ASR and speech translation (IBM) | ~4.6 GB |
Granite-Speech-5.0-2B-GGUF | ASR and speech translation (IBM, Q8_0 language model) | ~3.1 GB |
Nemotron-3.5-Transcribe-30B-A4B-FP8 | ASR, speech translation, speaker labels, word timestamps and single-model diarization (NVIDIA; needs about 32 GiB of memory) | ~34.7 GB |
Voxtral-Mini-4B-Realtime-2602 | ASR (offline transcription; realtime planned) | ~8 GB |
LFM3-Audio-2B-GGUF | TTS, ASR and voice chat (two voices, 19 TTS locales, 18 ASR languages); downloads the language model and mmproj | ~1.46 GB |
LFM3-Audio-2B-Q8_0-GGUF | LFM3 Audio with a Q8_0 language model | ~2.20 GB |
Qwen3.5-4B | Chat | ~3.4 GB |
LFM2.5-1.2B-Instruct-GGUF | Chat | ~0.7 GB |
LFM3-1.7B-GGUF | Chat (reasoning, tools) | ~1.1 GB |