Ultra-fast speech transcription service that converts audio files to text in record time - 1 hour of audio transcribes in ~20 seconds.

Quick Start

# Basic transcription (auto-detect language and dialect)
python3 scripts/transcribe.py /path/to/audio.mp3
# Save to file
python3 scripts/transcribe.py /path/to/audio.wav --output result.txt
# With domain-specific optimization
python3 scripts/transcribe.py /path/to/audio.mp3 --pd medical
# With speaker separation
python3 scripts/transcribe.py /path/to/meeting.mp3 --vspp-on 1 --speaker-num 2

Setup

1. API Credentials

Get credentials from iFlytek Open Platform:

• APP_ID: Application ID
• API_KEY: API key for authentication
• API_SECRET: API secret for signing requests

2. Environment Variables

export XFEI_APP_ID="your_app_id"
export XFEI_API_KEY="your_api_key"
export XFEI_API_SECRET="your_api_secret"

API Parameters

Required Parameters

Optional Parameters

Audio Requirements

• Format: MP3
• Sample rate: 16kHz
• Bit depth: 16-bit
• Channels: Mono (single channel)
• Size: ≤ 500MB
• Duration: ≤ 5 hours (recommended: ≥ 5 minutes)

Workflow

1. Upload Audio File

Files < 30MB use direct upload. Files ≥ 30MB use multipart upload (5MB chunks).

2. Create Transcription Task

Submit uploaded file URL with transcription parameters.

3. Poll for Results

Query task status periodically until completion.

Response Format

{
  "task_id": "1568100557463963551003",
  "task_status": "4",
  "text": "Transcribed text content...",
  "segments": [
    {
      "speaker": "spk-0",
      "begin": "0",
      "end": "470",
      "text": "听说。"
    }
  ]
}

Task Status

• 1: Pending
• 2: Processing
• 3: Completed
• 4: Callback completed
• -1: Failed

Language Support

autodialect (language=zh_cn)

• Major: Mandarin, Cantonese, Taiwanese, Sichuanese, Shanghainese, Northeastern
• Full list: 合肥话、芜湖话、皖北话、粤语、北京话、福州话、闽南语、潮汕话、客家话、贵阳话、海口话、石家庄话、太原话、郑州话、东北话、武汉话、长沙话、南京话、南昌话、大连话、呼和浩特话、银川话、西宁话、济南话、西安话、上海话、四川话、台湾话、天津话、乌鲁木齐话、云南话、杭州话、重庆话 (202 total)

Common Use Cases

• Meeting Transcription: Convert meeting recordings to text with speaker separation
• Interview Recording: Transcribe interviews for documentation
• Lecture Recording: Convert academic lectures to searchable text
• Voice Notes: Transform voice memos into text notes
• Call Center: Analyze customer service calls
• Legal Proceedings: Transcribe court hearings with domain optimization
• Medical Consultation: Doctor-patient conversation documentation

Error Handling

• 📖 接口文档：https://console.xfyun.cn/services/ost
• 💰 购买套餐：https://www.xfyun.cn/services/fast_lfasr?target=price

常见问题 FAQ

Q: 录音文件转写极速版的主要功能是什么？ A: 快速地将长段音频（5小时以内）数据转换成文本数据呢～ (๑•̀ㅂ•́)و✧

Q: 录音文件转写极速版支持什么语言？ A: 支持中文、英文 + 202种方言免切识别哦！ ヽ(✿ﾟ▽ﾟ)ノ

Q: 录音文件转写极速版支持什么应用平台？ A: 目前支持 WebAPI 应用平台啦～

Q: 为什么只支持 MP3 格式呀？ A: 因为 MP3 格式兼容性好、文件小、传输快呢～ 使用 lame 编码就能轻松接入啦！ (◕‿◕)

Tips

• For speaker separation: Use --vspp-on 1 for better speaker diarization
• For specific domains: Use --pd parameter for improved accuracy
• For faster processing: Audio files ≥ 5 minutes are prioritized
• For subtitle output: Use --enable-subtitle 1 for subtitle-formatted output
• For hot words: Use --dhw="word1,word2" to boost recognition accuracy