ElevenLabs声音转换替代:API语音输入与台词复核

apizt 实测记录 · 2026-10-24 · 不是第三方产品界面截图

通过 apizt 的 gpt-audio-2025-08-28 实际输入原创合成语音,得到新的 WAV。另一次 Whisper 请求识别出原台词,但原音约7.17秒、结果约9.3秒,不能宣称时间与停顿完全保留。这是语音到语音替代示范,不是 ElevenLabs 原生 Voice Changer。

本次操作

  1. 准备有权使用的源音频:本例为系统TTS合成教学台词,不是真人音色,也没有模仿特定人物。
  2. 使用支持音频的当前型号:chat/completions 中传 input_audio,modalities包含text和audio;实际旧型号503,新型号返回WAV。
  3. 独立核对输出:下载WAV,检测时长、声道和解码;使用Whisper独立转写,比较台词与停顿,保留速度变化。

结果与限制

speech-to-speech-current

httpmodelrequestSeconds
200gpt-audio-2025-08-289.476

output.wav

asr.json

{
  "task": "transcribe",
  "language": "english",
  "duration": 9.300000190734863,
  "text": "This is an original VoiceChanger practice recording. The map is ready. Please meet me at the North Gate.",
  "words": [
    {
      "word": "This",
      "start": 0.0,
      "end": 0.3199999928474426
    },
    {
      "word": "is",
      "start": 0.3199999928474426,
      "end": 0.6200000047683716
    },
    {
      "word": "an",
      "start": 0.6200000047683716,
      "end": 0.8600000143051147
    },
    {
      "word": "original",
      "start": 0.8600000143051147,
      "end": 1.1799999475479126
    },
    {
      "word": "VoiceChanger",
      "start": 1.1799999475479126,
      "end": 1.9199999570846558
    },
    {
      "word": "practice",
      "start": 1.9199999570846558,
      "end": 2.4600000381469727
    },
    {
      "word": "recording",
      "start": 2.4600000381469727,
      "end": 3.0199999809265137
    },
    {
      "word": "The",
      "start": 4.0,
      "end": 4.599999904632568
    },
    {
      "word": "map",
      "start": 4.599999904632568,
      "end": 4.880000114440918
    },
    {
      "word": "is",
      "start": 4.880000114440918,
      "end": 5.239999771118164
    },
    {
      "word": "ready",
      "start": 5.239999771118164,
      "end": 5.460000038146973
    },
    {
      "word": "Please",
      "start": 7.21999979019165,
      "end": 7.519999980926514
    },
    {
      "word": "meet",
      "start": 7.519999980926514,
      "end": 7.739999771118164
    },
    {
      "word": "me",
      "start": 7.739999771118164,
      "end": 8.0
    },
    {
      "word": "at",
      "start": 8.0,
      "end": 8.199999809265137
    },
    {
      "word": "the",
      "start": 8.199999809265137,
      "end": 8.5
    },
    {
      "word": "North",
      "start": 8.5,
      "end": 8.65999984741211
    },
    {
      "word": "Gate",
      "start": 8.65999984741211,
      "end": 8.9399995803833
    }
  ],
  "usage": {
    "type": "duration",
    "seconds": 10
  }
}
查看原始模型输出

speech-to-speech-api

httpmodelrequestSecondserror
503gpt-4o-audio-preview-2024-12-174.55{"message": "The model service is temporarily unavailable. Please try again later. (request id: 20261024020005109178074hSiU88ze)", "type": "new_api_error", "param": "", "code": ""}

复跑与凭据

下载本题 evidence.zip,保留各任务目录。API脚本通过本机环境变量 tokenba_API_KEY 读取你的Key;不把Key放进请求JSON。先核对模型广场当前可用型号和价格,再显式运行一次小样。

生成次数、文件成功、内容质检分别判断。没有在公开报告中保存任何可用Key。