技术博客：深入解析Amazon Polly——基于AWS的高性能神经文本转语音服务

# 中国区endpoint需指定--region和--endpoint-url
aws polly synthesize-speech \
  --region cn-north-1 \
  --endpoint-url https://polly.cn-north-1.amazonaws.com.cn \
  --output-format mp3 \
  --voice-id Zhiyu \
  --text "欢迎使用Amazon Polly神经文本转语音服务" \
  welcome.mp3

import boto3
 
# 初始化中国区Polly客户端
polly = boto3.client(
    'polly',
    region_name='cn-north-1',
    endpoint_url='https://polly.cn-north-1.amazonaws.com.cn'
)
 
# SSML文本
ssml_text = """
<speak>
  今天的最高气温是<prosody rate="slow">30摄氏度</prosody>，<break strength="medium"/>
  请注意防暑降温。<phoneme alphabet="ipa" ph="/ˈpɒli/">Polly</phoneme>支持专业术语的自定义发音。
</speak>
"""
 
# 合成语音
response = polly.synthesize_speech(
    Text=ssml_text,
    OutputFormat='mp3',
    VoiceId='Zhiyu',
    TextType='ssml'  # 必须指定为ssml
)
 
# 保存音频文件
with open('ssml_example.mp3', 'wb') as f:
    f.write(response['AudioStream'].read())
print("SSML定制语音生成完成！")

import boto3
import time
 
polly = boto3.client('polly', region_name='cn-north-1', endpoint_url='https://polly.cn-north-1.amazonaws.com.cn')
 
# 长文本（支持10万字以内）
long_text = """
这是一篇适合有声书的长文本示例。Amazon Polly的长文本神经语音可以无需分割文本，直接生成流畅的音频。...（省略剩余文本）
"""
 
# 启动异步合成任务
response = polly.start_speech_synthesis_task(
    Text=long_text,
    OutputFormat='mp3',
    VoiceId='Zhiyu',
    OutputS3BucketName='your-s3-bucket-name',  # 替换为你的S3桶名
    OutputS3KeyPrefix='polly-audio/'
)
 
task_id = response['SynthesisTask']['TaskId']
print(f"任务启动成功，ID：{task_id}")
 
# 轮询任务状态
while True:
    task = polly.get_speech_synthesis_task(TaskId=task_id)
    status = task['SynthesisTask']['TaskStatus']
    if status in ['completed', 'failed']:
        break
    print(f"任务状态：{status}，等待中...")
    time.sleep(10)
 
if status == 'completed':
    print(f"合成完成，音频地址：{task['SynthesisTask']['OutputUri']}")
else:
    print(f"合成失败：{task['SynthesisTask']['TaskStatusReason']}")

<lexicon version="1.0" 
         xmlns="http://www.w3.org/2005/01/pronunciation-lexicon"
         xml:lang="zh-CN"
         alphabet="ipa">
  <lexeme>
    <grapheme>豆包</grapheme>
    <phoneme>/dòu bāo/</phoneme>
  </lexeme>
  <lexeme>
    <grapheme>AI</grapheme>
    <phoneme>/rén gōng zhì néng/</phoneme>
  </lexeme>
</lexicon>

aws polly put-lexicon \
  --region cn-north-1 \
  --name DoubaoLexicon \
  --content file://doubao_lexicon.xml

aws polly synthesize-speech \
  --region cn-north-1 \
  --voice-id Zhiyu \
  --text "豆包是AI领域的优秀产品" \
  --lexicon-names DoubaoLexicon \
  custom_lexicon.mp3