audio_codebuddy/deploy-package/scripts/compare-aliyun-tts.sh

#!/bin/bash
set -e

B_URL='http://dashscope-a717.oss-cn-beijing.aliyuncs.com/1d/c2/20260712/9f107364/2b785586-afa2-445d-b4b3-5bb3a9939781.wav?Expires=1783879166&OSSAccessKeyId=LTAI5tPxpiCM2hjmWrFXrym1&Signature=cw%2BT4is4ceG0uxc7k7IWE8O9e0c%3D'
curl -sS -m 30 -o /tmp/aliyun-B-slow.wav "$B_URL"
ls -la /tmp/aliyun-B-slow.wav
ffmpeg -i /tmp/aliyun-B-slow.wav -f null - 2>&1 | grep -E 'Duration|Stream' | head -2

echo
echo '=== C: A + ffmpeg atempo=0.78 后处理(模拟播放器减速) ==='
ffmpeg -i /tmp/aliyun-A.mp3 -filter:a 'atempo=0.78' -y /tmp/aliyun-C-atempo.mp3 2>&1 | tail -1
ffmpeg -i /tmp/aliyun-C-atempo.mp3 -f null - 2>&1 | grep Duration | head -1

echo
echo '=== 对比 ==='
A_SEC=$(ffmpeg -i /tmp/aliyun-A.mp3 -f null - 2>&1 | grep Duration | head -1 | sed 's/Duration: //; s/,.*//' | awk -F: '{print ($1*3600+$2*60+$3)}')
B_SEC=$(ffmpeg -i /tmp/aliyun-B-slow.wav -f null - 2>&1 | grep Duration | head -1 | sed 's/Duration: //; s/,.*//' | awk -F: '{print ($1*3600+$2*60+$3)}')
C_SEC=$(ffmpeg -i /tmp/aliyun-C-atempo.mp3 -f null - 2>&1 | grep Duration | head -1 | sed 's/Duration: //; s/,.*//' | awk -F: '{print ($1*3600+$2*60+$3)}')

python3 - <<EOF
A=$A_SEC; B=$B_SEC; C=$C_SEC
print(f"A 阿里云 1.0x 默认:        {A:.2f}s")
print(f"B 阿里云 Instruct 语速较慢: {B:.2f}s  ({B/A:.3f}x)")
print(f"C 阿里云 1.0x + atempo=0.78:{C:.2f}s  ({C/A:.3f}x)")
print()
print(f"B vs C 差异: {abs(B-C)/A*100:.2f}%")
if abs(B-C)/A < 0.05:
    print("=> 结论: B ≈ C, 阿里云 Instruct 语速控制也是后处理变速(类似 Edge TTS)")
else:
    print("=> 结论: B 与 C 显著不同, 阿里云 Instruct 是真正的模型级慢速生成")
EOF

echo
echo '=== 顺便看看波形是不是真不同(语谱图) ==='
ffmpeg -i /tmp/aliyun-A.mp3 -lavfi showspectrumpic=s=1024x512 /tmp/aliyun-A-spec.png 2>&1 | tail -1
ffmpeg -i /tmp/aliyun-B-slow.wav -lavfi showspectrumpic=s=1024x512 /tmp/aliyun-B-spec.png 2>&1 | tail -1
ffmpeg -i /tmp/aliyun-C-atempo.mp3 -lavfi showspectrumpic=s=1024x512 /tmp/aliyun-C-spec.png 2>&1 | tail -1
ls -la /tmp/aliyun-*-spec.png

echo
echo '=== 把三个 mp3 上传到 OSS 让你试听 ==='
cp /data/ai/audio/server/.env /tmp/dashscope.env 2>/dev/null
cd /data/ai/audio/server && node -e '
const OSS = require("ali-oss").default || require("ali-oss");
require("dotenv").config({ path: "/data/ai/audio/server/.env" });
const c = {
  region: "oss-cn-shanghai",
  accessKeyId: process.env.OSS_ACCESS_KEY_ID,
  accessKeySecret: process.env.OSS_ACCESS_KEY_SECRET,
  bucket: process.env.OSS_BUCKET_NAME,
  internal: true,
  endpoint: process.env.OSS_INTERNAL_ENDPOINT,
};
const client = new OSS(c);
const fs = require("fs");
const files = {
  "aliyun-A-default": "/tmp/aliyun-A.mp3",
  "aliyun-B-instruct-slow": "/tmp/aliyun-B-slow.wav",
  "aliyun-C-atempo-postprocess": "/tmp/aliyun-C-atempo.mp3",
};
(async () => {
  for (const [name, path] of Object.entries(files)) {
    const key = `audio/voice-samples/aliyun-${name}.mp3`;
    await client.put(key, fs.readFileSync(path));
    console.log("uploaded:", key);
  }
  const PUB = process.env.OSS_ENDPOINT;
  for (const name of Object.keys(files)) {
    console.log(`https://${process.env.OSS_BUCKET_NAME}.${PUB}/audio/voice-samples/aliyun-${name}.mp3`);
  }
})().catch(e => { console.error(e); process.exit(1); });
'