pip install qwen3-tts

git clone https://github.com/QwenLM/Qwen3-TTS
cd Qwen3-TTS
pip install -e .

from qwen3_tts import Qwen3TTS
 
# Initialize the model
model = Qwen3TTS.from_pretrained("Qwen/Qwen3-TTS")
 
# Generate speech
audio = model.synthesize(
    text="Hello, welcome to the future of voice AI.",
    language="en"
)
 
# Save to file
audio.save("output.wav")

# Chinese
audio_zh = model.synthesize(
    text="欢迎使用语音人工智能",
    language="zh"
)
 
# Spanish
audio_es = model.synthesize(
    text="Bienvenido al futuro de la voz AI",
    language="es"
)
 
# Japanese
audio_ja = model.synthesize(
    text="音声AIの未来へようこそ",
    language="ja"
)

# Clone from reference audio
cloned_voice = model.clone_voice(
    reference_audio="speaker_sample.wav",
    reference_text="This is my voice sample."
)
 
# Use the cloned voice
audio = model.synthesize(
    text="Now I can speak in a cloned voice.",
    voice=cloned_voice
)

# Stream audio chunks
for chunk in model.synthesize_stream(
    text="This is a longer piece of text that will be streamed.",
    language="en"
):
    # Process each audio chunk
    play_audio(chunk)

audio = model.synthesize(
    text="I'm so excited to share this news with you!",
    language="en",
    emotion="excited",
    speed=1.1  # Slightly faster
)

FROM nvidia/cuda:12.1-runtime-ubuntu22.04
 
WORKDIR /app
COPY requirements.txt .
RUN pip install -r requirements.txt
 
COPY . .
CMD ["python", "server.py"]

from fastapi import FastAPI
from qwen3_tts import Qwen3TTS
 
app = FastAPI()
model = Qwen3TTS.from_pretrained("Qwen/Qwen3-TTS")
 
@app.post("/synthesize")
async def synthesize(text: str, language: str = "en"):
    audio = model.synthesize(text=text, language=language)
    return StreamingResponse(audio.to_wav(), media_type="audio/wav")

texts = [
    "First sentence to synthesize.",
    "Second sentence to synthesize.",
    "Third sentence to synthesize."
]
 
# Process in batch for better throughput
audios = model.synthesize_batch(texts, language="en")

model = Qwen3TTS.from_pretrained(
    "Qwen/Qwen3-TTS",
    quantization="int8"  # or "int4" for more aggressive quantization
)