ifly-voiceclone-tts

Clone a voice from audio samples and synthesize speech with it, using iFlytek's Voice Clone (声音复刻) API. Two-phase workflow: train a voice model, then synthesize speech with it.

Setup

1. Create an app at 讯飞控制台 with 一句话声音复刻 service enabled
2. Set environment variables:

   export IFLY_APP_ID="your_app_id"
   export IFLY_API_KEY="your_api_key"
   export IFLY_API_SECRET="your_api_secret"
   

Workflow

Phase 1: Train a Voice Model

Step 1 — Get training text

python3 scripts/voiceclone.py train get-text

This returns a list of text segments with segId. You need to record yourself reading one of these texts.

Step 2 — Create a training task

python3 scripts/voiceclone.py train create --name "MyVoice" --sex female --engine omni_v1

Returns task_id. Supported engines:

• omni_v1 — Multi-style universal voice (recommended)

Gender: male/female (or 1/2).

Step 3 — Upload audio

# Local file:
python3 scripts/voiceclone.py train upload --task-id 12345 --audio recording.wav --text-id 5001 --seg-id 1

# URL:
python3 scripts/voiceclone.py train upload --task-id 12345 --audio-url "https://example.com/voice.wav" --text-id 5001 --seg-id 1

Audio requirements:

• Format: WAV/MP3/M4A/PCM
• Duration: match the training text (typically 3-60 seconds)
• Quality: clear recording, minimal background noise

Step 4 — Submit for training

python3 scripts/voiceclone.py train submit --task-id 12345

Step 5 — Check status (poll until done)

python3 scripts/voiceclone.py train status --task-id 12345

When complete, returns the res_id (voice resource ID) needed for synthesis.

Quick one-shot training

python3 scripts/voiceclone.py train quick \
    --audio recording.wav \
    --name "MyVoice" \
    --sex female \
    --wait

This combines create → upload → submit → poll in one command. --wait polls every 30s until training completes and prints the res_id.

Phase 2: Synthesize Speech

# Basic synthesis
python3 scripts/voiceclone.py synth "你好，这是我的声音克隆。" --res-id YOUR_RES_ID

# With output file
python3 scripts/voiceclone.py synth "Hello world" --res-id YOUR_RES_ID --output hello.mp3

# From file
python3 scripts/voiceclone.py synth --file article.txt --res-id YOUR_RES_ID -o article.mp3

# From stdin
echo "测试语音合成" | python3 scripts/voiceclone.py synth --res-id YOUR_RES_ID

# Adjust parameters
python3 scripts/voiceclone.py synth "快一点" --res-id YOUR_RES_ID --speed 70 --volume 80

Train Subcommands

Synthesis Options

Notes

• Training API: HTTP REST at http://opentrain.xfyousheng.com/voice_train (MD5-based token auth)
• Synthesis API: WebSocket at wss://cn-huabei-1.xf-yun.com/v1/private/voice_clone (HMAC-SHA256 URL auth)
• vcn: always x6_clone for cloned voice synthesis
• Engine omni_v1: multi-style universal voice, supports cn/en/jp/ko/ru
• Training text: use get-text to find available text segments — you must record yourself reading the corresponding text
• Training time: typically 2–10 minutes depending on load
• No pip dependencies: uses pure Python stdlib (built-in WebSocket client)
• Env vars: IFLY_APP_ID, IFLY_API_KEY, IFLY_API_SECRET
• Output: prints absolute path of saved audio to stdout
• API doc: https://www.xfyun.cn/doc/spark/voiceclone.html

常见错误码速查指南 ฅ⁽͑˙˙⁾ฅ

遇到错误先别慌～看看下面的错误码对照表就知道怎么办啦 ✧｡･ﾟ:*･

🎤 音色训练接口 - 常见错误码

💡 小贴士：如果是权限、授权相关的问题（10016-10021），基本上都需要联系讯飞官方处理哦～可以提交工单：https://console.xfyun.cn/workorder/commit

🎵 音频合成接口 - 常见错误码

💡 超重要！ 错误码100001-100010可能是引擎层面的问题，提交工单时记得提供：

• sid（请求会话ID）
• 完整的错误信息
• 复现步骤

这样技术人员才能快速帮你定位问题哦～ ✧٩(ˊᗜˋ*)و

🆘 遇到问题怎么办？

1. 先看错误码：上面的表格基本上涵盖了常见错误，看看有没有对应的～ ๑•̀ㅂ•́)و✧
2. 检查参数：很多错误都是参数写错导致的，对照接口文档仔细核对一下哦！
3. 提交工单：如果表格里没有，或者搞不定，点击这里提交工单：https://console.xfyun.cn/workorder/commit
4. 购买/升级服务：需要更多调用量或功能的话：
   • 一句话声音复刻控制台
   • 购买服务包

🎉 祝你开发顺利！ 如果有其他问题也可以随时问我哦～ 一起加油！(´▽`ʃ♡ƪ)