Streaming Generation

import torch
import soundfile as sf
from qwen_tts import Qwen3TTSModel

model = Qwen3TTSModel.from_pretrained(
    "Qwen/Qwen3-TTS-12Hz-1.7B-CustomVoice",
    device_map="cuda:0",
    dtype=torch.bfloat16,
    attn_implementation="flash_attention_2",
)

# Streaming generation
wavs, sr = model.generate_custom_voice(
    text="Hello, this is a streaming generation test.",
    language="English",
    speaker="Ryan",
    non_streaming_mode=False,  # Enable streaming
)

sf.write("output.wav", wavs[0], sr)

import torch
from qwen_tts import Qwen3TTSModel

model = Qwen3TTSModel.from_pretrained(
    "Qwen/Qwen3-TTS-12Hz-1.7B-CustomVoice",
    device_map="cuda:0",
    dtype=torch.bfloat16,
    attn_implementation="flash_attention_2",
)

# Low-latency streaming
wavs, sr = model.generate_custom_voice(
    text="Welcome to our customer service. How may I help you today?",
    language="English",
    speaker="Ryan",
    non_streaming_mode=False,
)

import torch
from qwen_tts import Qwen3TTSModel

model = Qwen3TTSModel.from_pretrained(
    "Qwen/Qwen3-TTS-12Hz-1.7B-VoiceDesign",
    device_map="cuda:0",
    dtype=torch.bfloat16,
    attn_implementation="flash_attention_2",
)

wavs, sr = model.generate_voice_design(
    text="This is a real-time voice generation test.",
    language="English",
    instruct="Male, professional news anchor, clear and authoritative",
    non_streaming_mode=False,
)

import torch
from qwen_tts import Qwen3TTSModel

model = Qwen3TTSModel.from_pretrained(
    "Qwen/Qwen3-TTS-12Hz-1.7B-Base",
    device_map="cuda:0",
    dtype=torch.bfloat16,
    attn_implementation="flash_attention_2",
)

ref_audio = "https://qianwen-res.oss-cn-beijing.aliyuncs.com/Qwen3-TTS-Repo/clone.wav"
ref_text = "Okay. Yeah. I resent you. I love you. I respect you. But you know what? You blew it! And thanks to you."

wavs, sr = model.generate_voice_clone(
    text="This message is being generated in real-time with minimal latency.",
    language="English",
    ref_audio=ref_audio,
    ref_text=ref_text,
    non_streaming_mode=False,
)

import torch
from qwen_tts import Qwen3TTSModel
import sounddevice as sd

model = Qwen3TTSModel.from_pretrained(
    "Qwen/Qwen3-TTS-12Hz-1.7B-CustomVoice",
    device_map="cuda:0",
    dtype=torch.bfloat16,
    attn_implementation="flash_attention_2",
)

def speak(text: str):
    """Generate and play speech with low latency"""
    wavs, sr = model.generate_custom_voice(
        text=text,
        language="Auto",
        speaker="Ryan",
        non_streaming_mode=False,  # Streaming for low latency
    )
    
    # Play audio immediately
    sd.play(wavs[0], sr)
    sd.wait()

# Real-time responses
speak("Hello! How can I assist you today?")
speak("I'm processing your request now.")
speak("Here are the results you requested.")

import torch
from qwen_tts import Qwen3TTSModel
from typing import List

model = Qwen3TTSModel.from_pretrained(
    "Qwen/Qwen3-TTS-12Hz-1.7B-CustomVoice",
    device_map="cuda:0",
    dtype=torch.bfloat16,
    attn_implementation="flash_attention_2",
)

def generate_commentary(events: List[str]):
    """Generate live commentary with minimal delay"""
    for event in events:
        wavs, sr = model.generate_custom_voice(
            text=event,
            language="English",
            speaker="Aiden",
            instruct="Excited sports commentator style",
            non_streaming_mode=False,
        )
        # Play immediately as each is generated
        play_audio(wavs[0], sr)

events = [
    "And here comes the player with the ball!",
    "What an incredible move!",
    "The crowd is going wild!",
]

generate_commentary(events)

import torch
from qwen_tts import Qwen3TTSModel

model = Qwen3TTSModel.from_pretrained(
    "Qwen/Qwen3-TTS-12Hz-1.7B-CustomVoice",
    device_map="cuda:0",
    dtype=torch.bfloat16,
    attn_implementation="flash_attention_2",
)

def ivr_prompt(message: str):
    """Generate IVR prompts with low latency"""
    wavs, sr = model.generate_custom_voice(
        text=message,
        language="English",
        speaker="Ryan",
        instruct="Clear, professional phone system voice",
        non_streaming_mode=False,
    )
    return wavs[0], sr

# Generate IVR prompts
greeting = ivr_prompt("Thank you for calling. Please select from the following options.")
option1 = ivr_prompt("Press 1 for customer service.")
option2 = ivr_prompt("Press 2 for technical support.")

# Ultra-low latency: Use 0.6B model
model = Qwen3TTSModel.from_pretrained(
    "Qwen/Qwen3-TTS-12Hz-0.6B-CustomVoice",  # Smaller, faster
    device_map="cuda:0",
    dtype=torch.bfloat16,
)

# Better quality, slightly higher latency: Use 1.7B model
model = Qwen3TTSModel.from_pretrained(
    "Qwen/Qwen3-TTS-12Hz-1.7B-CustomVoice",  # Larger, higher quality
    device_map="cuda:0",
    dtype=torch.bfloat16,
)

# Use FlashAttention-2 for best performance
model = Qwen3TTSModel.from_pretrained(
    "Qwen/Qwen3-TTS-12Hz-1.7B-CustomVoice",
    device_map="cuda:0",
    dtype=torch.bfloat16,
    attn_implementation="flash_attention_2",  # Critical for performance
)

# Optimize for speed
wavs, sr = model.generate_custom_voice(
    text="Your text",
    language="English",
    speaker="Ryan",
    non_streaming_mode=False,
    max_new_tokens=2048,     # Limit max length
    temperature=0.7,         # Lower for faster, more deterministic
    top_k=20,                # Reduce for speed
)

from dashscope import SpeechSynthesizer

# Real-time streaming API
response = SpeechSynthesizer.call(
    model='qwen3-tts',
    text='Your text here',
    format='wav',
    sample_rate=24000,
    streaming=True,  # Enable streaming
)

# Process chunks as they arrive
for chunk in response:
    play_audio_chunk(chunk)