Flat model repositories accept an alias or a full repository ID. tts.load("fish-s2-pro") and tts.load("appautomaton/fishaudio-s2-pro-8bit-mlx") are equivalent. For a repository containing multiple artifacts, use an alias or specify artifact_subdir. Original checkpoint paths work where the model guide documents their layout.

¹ DramaBox also downloads the Gemma 3 12B text encoder automatically. Its optional denoise_ref=True setting uses the MLX RE-USE / SEMamba enhancer to clean noisy voice references. Denoising is off by default, and the enhancer weights carry the NSCLv1 non-commercial license. The DramaBox guide covers these components and advanced controls.

² Breeze TTS 2 has no published MLX weights or alias yet. Download the official checkpoint, convert it with scripts/convert/breeze_tts.py, and pass the converted directory to tts.load(). The weights and outputs made from them carry the BreezeBlue non-commercial license. The Breeze TTS 2 guide covers conversion, controls, and known limits.

Voice cloning with emotion tags

import mlx_speech
from mlx_speech.audio import write_wav

model = mlx_speech.tts.load("fish-s2-pro")
result = model.generate(
    "[excited] This is amazing!",
    reference_audio="reference.wav",
    reference_text="Transcript of the reference audio.",
)
write_wav("cloned.wav", result.waveform, sample_rate=result.sample_rate)

Live transcription with Confucius4-R2T2

import mlx_speech

asr = mlx_speech.asr.load("confucius4-r2t2")
session = asr.stream_session(language="English")
for pcm in microphone:                 # float32, 16 kHz mono
    print(session.feed(pcm).committed)
print(session.finalize().text)

Streaming transcription with Nemotron

import mlx_speech
from mlx_speech.audio import load_audio

nemotron = mlx_speech.asr.load("nemotron-asr-streaming")
session = nemotron.stream_session(language="en-US", att_context_size=(56, 3))
waveform, _ = load_audio("audio.wav", sample_rate=16_000, mono=True)
for start in range(0, int(waveform.size), 1_600):
    session.feed(waveform[start : start + 1_600])
session.finalize()
print(session.result().text)

Granite transcription and model discovery

import mlx_speech

# Granite defaults to the published selective-int8 artifact
granite = mlx_speech.asr.load("granite-speech-4.0-1b")
print(granite.generate("audio.wav").text)

# Discover models
mlx_speech.tts.list_models()
mlx_speech.tts.list_models(detailed=True)  # includes shared-repo artifact paths
mlx_speech.asr.list_models()

# Bounded waveform streaming with dots.tts
mlx-speech tts --model dots-tts-soar --text "Hello!" --stream -o streamed.wav

# Voice cloning with emotion tags
mlx-speech tts --model fish-s2-pro \
  --text "[whisper] Just between us..." \
  --reference-audio ref.wav \
  --reference-text "Transcript of reference." \
  -o cloned.wav

# Step Audio emotion editing
mlx-speech tts --model step-audio \
  --reference-audio input.wav \
  --reference-text "Transcript." \
  --edit-type emotion --edit-info happy \
  -o happy.wav

# Sound effect generation
mlx-speech tts --model moss-sound-effect \
  --text "rolling thunder with rainfall" \
  --duration-seconds 8 \
  -o thunder.wav

# Transcribe audio with any ASR selector
mlx-speech asr --model cohere-asr --audio speech.wav
mlx-speech asr --model qwen3-asr-1.7b --audio speech.wav --language Chinese

# Local checkpoint paths work anywhere an alias does
mlx-speech tts --model models/fish_s2_pro/mlx-int8 --text "Hello!" -o output.wav

# Discover models
mlx-speech tts --list-models
mlx-speech asr --list-models