AgentSkillsCN

speech2text-client

利用 Speech2Text（语音转文字）服务对音频/视频进行转录：注册/登录（通过 x-jwt-token 使用 JWT），上传文件（/tasks/add），轮询任务状态（/tasks/{id}），获取文本结果（/results/txt），并排查常见错误。可通过诸如“speech2text”、“Speech2Text”、“语音识别”、“将音频/视频/语音转录成文字”、“从 tg_uploads 中获取转录稿”、“如何获取令牌/登录”、“API 的 curl 示例”等请求触发。

SKILL.md
--- frontmatter
name: speech2text-client
description: 'Use a Speech2Text (speech-to-text) service to transcribe audio/video: register/login (JWT via x-jwt-token), upload files (/tasks/add), poll status (/tasks/{id}), fetch text results (/results/txt), and troubleshoot common errors. Triggered by requests like "speech2text", "Speech2Text", "распознавание речи", "расшифруй/транскрибируй аудио/видео/голосовое", "получи транскрипцию из tg_uploads", "как получить токен/логин", "curl пример для API".'

speech2text-client

Quick start

  • •Swagger/OpenAPI: ${SPEECH2TEXT_BASE_URL}/openapi.json
  • •UI (optional): (depends on your deployment)
  • •CLI helper in this repo: python3 scripts/speech2text.py

Check connectivity:

bash
python3 scripts/speech2text.py doctor

Auth (register/login → token)

  • •API uses JWT in header x-jwt-token.
  • •Store token outside the repo (default): ~/.config/speech2text/token (mode 600).

If you already have an account:

bash
python3 scripts/speech2text.py login --username "YOUR_LOGIN"

If you don’t have an account yet (one-time):

bash
python3 scripts/speech2text.py register --user-name "YOUR_LOGIN"
python3 scripts/speech2text.py login --username "YOUR_LOGIN"

Alternative: provide token via env var (no file write):

bash
export SPEECH2TEXT_TOKEN="..."

Transcribe (no diarization, ≤ 5 min)

Defaults are optimized for single-speaker voice notes:

  • •diarization=false (default)
  • •max_speakers=1
  • •timeout is clamped to 300s

Transcribe a file:

bash
python3 scripts/speech2text.py transcribe path/to/voice.ogg --out tmp/voice.txt

If the service rejects the format, convert to WAV and retry:

bash
ffmpeg -y -i path/to/voice.ogg -ac 1 -ar 16000 tmp/voice.wav
python3 scripts/speech2text.py transcribe tmp/voice.wav --out tmp/voice.txt

Minimal curl cheatsheet

bash
# Upload (no diarization) — adjust base URL for your deployment
curl -sS -X POST \
  "$SPEECH2TEXT_BASE_URL/tasks/add?max_speakers=1&diarization=false&data_format=AUDIO" \
  -H "x-jwt-token: $SPEECH2TEXT_TOKEN" \
  -F "uploaded_file=@voice.ogg"

# Poll
curl -sS "$SPEECH2TEXT_BASE_URL/tasks/<uuid>" -H "x-jwt-token: $SPEECH2TEXT_TOKEN"

# Result
curl -sS "$SPEECH2TEXT_BASE_URL/results/txt?task_id=<uuid>" -H "x-jwt-token: $SPEECH2TEXT_TOKEN"

Notes

  • •For a Telegram voice note saved under tg_uploads/, pass that .ogg file path into transcribe.
  • •For multi-speaker meetings, enable diarization explicitly: --diarization --max-speakers N (slower).
  • •If you see CUDA failed with error out of memory, the service is likely overloaded — wait a bit and retry.