Bark 文字转语音¶

pip install git+https://github.com/suno-ai/bark.git

# 或通过 pip
pip install bark

# 环境变量配置
export SUNO_USE_SMALL_MODELS=True    # 使用小模型（节省显存）
export SUNO_OFFLOAD_CPU=True          # GPU 不足时卸载到 CPU

from bark import SAMPLE_RATE, generate_audio, preload_models
from scipy.io.wavfile import write as write_wav

# 预加载模型
preload_models()

# 生成语音
audio_array = generate_audio("Hello, this is a test of Bark text to speech.")

# 保存为 WAV
write_wav("output.wav", SAMPLE_RATE, audio_array)

# 中文语音预设
audio = generate_audio(
    "你好，我是一个人工智能语音助手。今天天气很好。",
    history_prompt="v2/zh_speaker_1"
)
write_wav("chinese_output.wav", SAMPLE_RATE, audio)

# 英文预设
audio = generate_audio("Hello world!", history_prompt="v2/en_speaker_3")

# 中文预设
audio = generate_audio("你好世界！", history_prompt="v2/zh_speaker_5")

# 日文预设
audio = generate_audio("こんにちは世界！", history_prompt="v2/ja_speaker_1")

# 可用语音预设：
# v2/en_speaker_0 ~ v2/en_speaker_9
# v2/zh_speaker_0 ~ v2/zh_speaker_9
# v2/ja_speaker_0 ~ v2/ja_speaker_9
# v2/de_speaker_0 ~ v2/de_speaker_9
# v2/fr_speaker_0 ~ v2/fr_speaker_9
# 等等...

# Bark 支持的特殊标记
texts = [
    "有时候... [laughs] 事情就是这么搞笑。",
    "[clears throat] 好了，我们开始正题。",
    "这是 [sighs] 很难的决定。",
    "♪ Do re mi fa sol la ti do ♪",  # 唱歌
    "— 等一下 —",                    # 停顿
]

for i, text in enumerate(texts):
    audio = generate_audio(text)
    write_wav(f"special_{i}.wav", SAMPLE_RATE, audio)

import numpy as np
from bark import generate_audio, SAMPLE_RATE
from scipy.io.wavfile import write as write_wav

def generate_long_audio(text, voice="v2/zh_speaker_3"):
    """将长文本分段生成然后拼接"""
    # 按句子分割
    sentences = text.replace("。", "。\n").replace("！", "！\n").replace("？", "？\n").split("\n")
    sentences = [s.strip() for s in sentences if s.strip()]

    audio_segments = []
    silence = np.zeros(int(0.3 * SAMPLE_RATE))  # 句间停顿

    for sentence in sentences:
        if len(sentence) > 200:  # 太长的句子再切分
            parts = [sentence[i:i+200] for i in range(0, len(sentence), 200)]
        else:
            parts = [sentence]

        for part in parts:
            audio = generate_audio(part, history_prompt=voice)
            audio_segments.append(audio)
            audio_segments.append(silence)

    # 拼接所有片段
    full_audio = np.concatenate(audio_segments)
    return full_audio

text = """
人工智能正在改变我们的生活。从自动驾驶到医疗诊断，AI的应用无处不在。
然而，我们也需要关注AI带来的伦理问题。隐私保护和算法公平性是重要的课题。
"""

audio = generate_long_audio(text)
write_wav("long_speech.wav", SAMPLE_RATE, audio)

from bark.generation import load_codec_model, generate_text_semantic
from bark.api import semantic_to_waveform
import numpy as np

# 注意：Bark 的语音克隆能力有限
# 需要用参考音频生成 history prompt

# 方法1：使用 generate_audio 的输出作为 prompt
from bark import generate_audio

# 先用特定语音生成一段参考
reference_audio = generate_audio(
    "This is my reference voice sample.",
    history_prompt="v2/en_speaker_6"
)

# 保存为自定义 prompt（高级用法需要深入源码）

from bark.generation import (
    generate_text_semantic,
    preload_models,
)
from bark.api import semantic_to_waveform
from bark import SAMPLE_RATE
import sounddevice as sd

preload_models()

def stream_speech(text, voice="v2/zh_speaker_3"):
    """逐句流式播放"""
    sentences = [s.strip() for s in text.split("。") if s.strip()]

    for sentence in sentences:
        sentence = sentence + "。"
        audio = generate_audio(sentence, history_prompt=voice)
        sd.play(audio, SAMPLE_RATE)
        sd.wait()

stream_speech("今天我们来讨论人工智能。机器学习是其中一个重要分支。深度学习推动了近年来的技术突破。")

from fastapi import FastAPI
from bark import generate_audio, SAMPLE_RATE, preload_models
from scipy.io.wavfile import write as write_wav
from fastapi.responses import FileResponse
import tempfile
import os

app = FastAPI()
preload_models()

@app.post("/tts")
async def text_to_speech(text: str, voice: str = "v2/zh_speaker_3"):
    audio = generate_audio(text, history_prompt=voice)

    tmp = tempfile.NamedTemporaryFile(suffix=".wav", delete=False)
    write_wav(tmp.name, SAMPLE_RATE, audio)

    return FileResponse(
        tmp.name,
        media_type="audio/wav",
        filename="speech.wav"
    )

import json
from bark import generate_audio, SAMPLE_RATE
from scipy.io.wavfile import write as write_wav
from pathlib import Path

# 批量配置
tasks = [
    {"text": "第一章：引言", "voice": "v2/zh_speaker_1", "output": "ch1_intro.wav"},
    {"text": "第二章：方法", "voice": "v2/zh_speaker_1", "output": "ch2_method.wav"},
    {"text": "第三章：结论", "voice": "v2/zh_speaker_1", "output": "ch3_conclusion.wav"},
]

output_dir = Path("./audiobook")
output_dir.mkdir(exist_ok=True)

for task in tasks:
    print(f"Generating: {task['output']}")
    audio = generate_audio(task["text"], history_prompt=task["voice"])
    write_wav(str(output_dir / task["output"]), SAMPLE_RATE, audio)

# 设置随机种子
import torch
torch.manual_seed(42)

audio = generate_audio("固定语音", history_prompt="v2/zh_speaker_3")